Default: `onDemand`.
Note: If month > 0 is set, `monthly` billing mode will be enforced. * `onDemand`: Pay-as-you-go billing * `monthly`: Monthly subscription billing * `spot`: Spot instance billing
Note: This parameter is only effective when monthly billing is enabled.
Note: This parameter is only effective when monthly billing is enabled and auto renew is enabled.
Enum: `instance`
Enum: `private`
Enum: `http`, `tcp`
Enum: `instance`
Enum: `private`
Enum: `http`, `tcp`
Enum: `official`, `community`, `private`
Note: * Query official/community templates: set to `false` * Query private templates: set to `true`
Enum: `png`, `webp`, `jpeg`
Enum: `TASK_STATUS_SUCCEED`, `TASK_STATUS_FAILED`
By default, the generated outputs will be saved to Novita AI Storage temporarily and privately.
If set to False, Novita AI will create an additional directory in the path to save the generated outputs. The default is False.
1: Explicit Nudity, Explicit Sexual Activity, Sex Toys; Hate Symbols; Non-Explicit Nudity, Obstructed Intimate Parts, Kissing on the Lips.
2: Explicit Nudity, Explicit Sexual Activity, Sex Toys; Hate Symbols; Non-Explicit Nudity, Obstructed Intimate Parts, Kissing on the Lips; Female Swimwear or Underwear, Male Swimwear or Underwear.
Enum: `0`, `1`, `2`
Enum: `Euler a`, `Euler`, `LMS`, `Heun`, `DPM2`, `DPM2 a`, `DPM++ 2S a`, `DPM++ 2M`, `DPM++ SDE`, `DPM fast`, `DPM adaptive`, `LMS Karras`, `DPM2 Karras`, `DPM2 a Karras`, `DPM++ 2S a Karras`, `DPM++ 2M Karras`, `DPM++ SDE Karras`, `DDIM`, `PLMS`, `UniPC`
Enum: `png, webp, jpeg`
Enum: `png`, `webp`, `jpeg`
Enum: `stable-diffusion-xl-base-1.0`, `dreamshaperXL09Alpha_alpha2Xl10_91562`, `protovisionXLHighFidelity3D_release0630Bakedvae_154359`, `v1-5-pruned-emaonly`, `epicrealism_naturalSin_121250`, `chilloutmix_NiPrunedFp32Fix`, `abyssorangemix3AOM3_aom3a3_10864`, `dreamshaper_8_93211`, `WFChild_v1.0`, `majichenmixrealistic_v10`, `realisticVisionV51_v51VAE_94301`, `sdxlUnstableDiffusers_v11_216694`, `realisticVisionV40_v40VAE_81510`, `epicrealismXL_v10_247189`, `somboy_v10_172675`, `yesmixXL_v10_283329`, `animagineXLV31_v31_325600`
Enum: `1e-4, 1e-5, 1e-6, 2e-4, 5e-5`
Enum: `constant, linear, cosine, cosine_with_restarts, polynomial, constant_with_warmup`
Enum: `face_crop_region`, `resize`, `face_restore`
Enum: `UNKNOWN`, `QUEUING`, `TRAINING`, `SUCCESS`, `CANCELED`, `FAILED`
Enum: `lora`
Enum: `DEPLOYING`, `SERVING`
Enum: `jpeg`, `png`, `webp`
Enum: `1`, `2`, `3`, `4`, `5`
Enum: `21:9`, `16:9`, `4:3`, `3:2`, `1:1`, `2:3`, `3:4`, `9:16`, `9:21`
Enum: `21:9`, `16:9`, `4:3`, `3:2`, `1:1`, `2:3`, `3:4`, `9:16`, `9:21`
The safety tolerance level for the generated image. 1 being the most strict and 5 being the most permissive. Default is 2.
Enum: `1`, `2`, `3`, `4`, `5`
Enum: `png` `webp` `jpeg`
Enum: `jpeg`, `png`, `webp`
Enum: `civitai`, `training`, `uploading`
Enum: `checkpoint`, `lora`, `vae`, `controlnet`, `upscaler`, `textualinversion`
Enum: `PUT`
Enum: `0`, `1`, `2`
Enum: `png`, `jpeg`
Enum: `DPM++ 2M Karras`, `DPM++ SDE Karras`, `DPM++ 2M SDE Exponential`, `DPM++ 2M SDE Karras`, `Euler a`, `Euler`, `LMS`, `Heun`, `DPM2`, `DPM2 a`, `DPM++ 2S a`, `DPM++ 2M`, `DPM++ SDE`, `DPM++ 2M SDE`, `DPM++ 2M SDE Heun`, `DPM++ 2M SDE Heun Karras`, `DPM++ 2M SDE Heun Exponential`, `DPM++ 3M SDE`, `DPM++ 3M SDE Karras`, `DPM++ 3M SDE Exponential`, `DPM fast`, `DPM adaptive`, `LMS Karras`, `DPM2 Karras`, `DPM2 a Karras`, `DPM++ 2S a Karras`, `Restart`, `DDIM`, `PLMS`, `UniPC`
Enum: `0`, `1`, `2`, `3`
Enum: `0`, `1`, `2`, `3`
Enum: `0`, `1`
Enum: `0`, `1`
Enum: `sd_xl_refiner_1.0.safetensors`
Enum: `none`, `canny`, `depth`, `depth_leres`, `depth_leres++`, `hed`, `hed_safe`, `mediapipe_face`, `mlsd`, `normal_map`, `openpose`, `openpose_hand`, `openpose_face`, `openpose_faceonly`, `openpose_full`, `clip_vision`, `color`, `pidinet`, `pidinet_safe`, `pidinet_sketch`, `pidinet_scribble`, `scribble_xdog`, `scribble_hed`, `segmentation`, `threshold`, `depth_zoe`, `normal_bae`, `oneformer_coco`, `oneformer_ade20k`, `lineart`, `lineart_coarse`, `lineart_anime`, `lineart_standard`, `shuffle`, `tile_resample`, `invert`, `lineart_anime_denoise`, `reference_only`, `reference_adain`, `reference_adain+attn`, `inpaint`, `inpaint_only`, `inpaint_only+lama`, `tile_colorfix`, `tile_colorfix+sharp`, `depth_anything`
Enum: `0`, `1`, `2`
Enum: `0`, `1`, `2`
Enum: `png`, `webp`, `jpeg`
Enum: `TASK_STATUS_SUCCEED`, `TASK_STATUS_FAILED`
By default, the generated outputs will be saved to Novita AI Storage temporarily and privately.
If set to False, Novita AI will create an additional directory in the path to save the generated outputs. The default is False.
1: Explicit Nudity, Explicit Sexual Activity, Sex Toys; Hate Symbols; Non-Explicit Nudity, Obstructed Intimate Parts, Kissing on the Lips.
2: Explicit Nudity, Explicit Sexual Activity, Sex Toys; Hate Symbols; Non-Explicit Nudity, Obstructed Intimate Parts, Kissing on the Lips; Female Swimwear or Underwear, Male Swimwear or Underwear. Enum: `0`, `1`, `2`
Enum: `Euler a`, `Euler`, `LMS`, `Heun`, `DPM2`, `DPM2 a`, `DPM++ 2S a`, `DPM++ 2M`, `DPM++ SDE`, `DPM fast`, `DPM adaptive`, `LMS Karras`, `DPM2 Karras`, `DPM2 a Karras`, `DPM++ 2S a Karras`, `DPM++ 2M Karras`, `DPM++ SDE Karras`, `DDIM`, `PLMS`, `UniPC`
Enum: `scribble_hed`, `softedge_hed`, `scribble_hedsafe`, `softedge_hedsafe`, `depth_midas`, `mlsd`, `openpose`, `openpose_face`, `openpose_faceonly`, `openpose_full`, `openpose_hand`, `dwpose`, `scribble_pidinet`, `softedge_pidinet`, `scribble_pidsafe`, `softedge_pidsafe`, `normal_bae`, `lineart_coarse`, `lineart_realistic`, `lineart_anime`, `lineart`, `depth_zoe`, `shuffle`, `mediapipe_face`, `canny`, `depth`, `depth_leres`, `depth_leres++`
Enum: `ip-adapter_sd15.bin`, `ip-adapter_sdxl.bin`
Enum: `png`, `webp`, `jpeg`
Enum: `TASK_STATUS_SUCCEED`, `TASK_STATUS_FAILED`
By default, the generated outputs will be saved to Novita AI Storage temporarily and privately.
If set to False, Novita AI will create an additional directory in the path to save the generated outputs. The default is False.
1: Explicit Nudity, Explicit Sexual Activity, Sex Toys; Hate Symbols; Non-Explicit Nudity, Obstructed Intimate Parts, Kissing on the Lips.
2: Explicit Nudity, Explicit Sexual Activity, Sex Toys; Hate Symbols; Non-Explicit Nudity, Obstructed Intimate Parts, Kissing on the Lips; Female Swimwear or Underwear, Male Swimwear or Underwear.
Enum: `0`, `1`, `2`
Enum: `Euler a`, `Euler`, `LMS`, `Heun`, `DPM2`, `DPM2 a`, `DPM++ 2S a`, `DPM++ 2M`, `DPM++ SDE`, `DPM fast`, `DPM adaptive`, `LMS Karras`, `DPM2 Karras`, `DPM2 a Karras`, `DPM++ 2S a Karras`, `DPM++ 2M Karras`, `DPM++ SDE Karras`, `DDIM`, `PLMS`, `UniPC`
Enum: `0`, `1`
Enum: `0`, `1`
Optional values: `5`, `10`
Optional values: `5`, `10`
Enum: `UNKNOWN`, `QUEUING`, `TRAINING`, `SUCCESS`, `CANCELED`, `FAILED`
Enum: `DEPLOYING`, `SERVING`
Option 1:
You can use the string type to represent the text contents of the message.
Option 2:
Use an array of content parts, object\[]. Detailed fields are as follows:
Only vision language models can be used.
An array of content parts, object\[]. Detailed fields are as follows:
Only models that support video can be used.
An array of content parts, object\[]. Detailed fields are as follows:
Only models that support audio can be used.
Output modality parameters:
An array of content parts, object\[]. Detailed fields are as follows:
Enum: `png`, `webp`, `jpeg`
Optional values: `6`, `10`
Optional values: `6`, `10`
Optional values: `6`, `10`
Optional values: `6`, `10`
- Chinese: moss\_audio\_ce44fc67-7ce3-11f0-8de5-96e35d26fb85, moss\_audio\_aaa1346a-7ce7-11f0-8e61-2e6e3c7ee85d, Chinese (Mandarin)\_Lyrical\_Voice, Chinese (Mandarin)\_HK\_Flight\_Attendant
- English: English\_Graceful\_Lady, English\_Insightful\_Speaker, English\_radiant\_girl, English\_Persuasive\_Man, moss\_audio\_6dc281eb-713c-11f0-a447-9613c873494c, moss\_audio\_570551b1-735c-11f0-b236-0adeeecad052, moss\_audio\_ad5baf92-735f-11f0-8263-fe5a2fe98ec8, English\_Lucky\_Robot
- Japanese: Japanese\_Whisper\_Belle, moss\_audio\_24875c4a-7be4-11f0-9359-4e72c55db738, moss\_audio\_7f4ee608-78ea-11f0-bb73-1e2a4cfcd245, moss\_audio\_c1a6a3ac-7be6-11f0-8e8e-36b92fbb4f95
- `0`: Success
- `1002`: Rate limit
- `1004`: Authentication failed
- `1039`: TPM rate limit triggered
- `1042`: Invalid characters exceed 10%
- `2013`: Parameter error
• Interjection tags: Only supported when model is `speech-2.8-hd` or `speech-2.8-turbo`. Supported interjections: `(laughs)` (laughter), `(chuckle)` (light laugh), `(coughs)` (cough), `(clear-throat)` (clear throat), `(groans)` (groan), `(breath)` (normal breathing), `(pant)` (panting), `(inhale)` (inhale), `(exhale)` (exhale), `(gasps)` (gasp), `(sniffs)` (sniff), `(sighs)` (sigh), `(snorts)` (snort), `(burps)` (burp), `(lip-smacking)` (lip smacking), `(humming)` (humming), `(hissing)` (hissing), `(emm)` (um), `(whistles)` (whistle), `(sneezes)` (sneeze), `(crying)` (crying), `(applause)` (applause)
• txt file: Length limit \<100000 characters. Supports custom pause using `<#x#>` tag. x is pause duration (in seconds), range \[0.01, 99.99], up to 2 decimal places. Pause must be set between two pronounceable text segments, cannot use multiple pause tags consecutively
• zip file:
• Compressed package must contain txt or json files of the same format.
• json file format: Supports \[`title`, `content`, `extra`] three fields, representing title, body, and additional information. If all three fields exist, 3 groups of results will be produced, 9 files in total, stored in one folder. If a field does not exist or is empty, no corresponding result will be generated
• The model will automatically match appropriate emotion based on input text, usually no need to specify manually\
• This parameter only works for `speech-2.6-hd`, `speech-2.6-turbo`, `speech-02-hd`, `speech-02-turbo`, `speech-01-hd`, `speech-01-turbo` models
• Options `fluent`, `whisper` only work for `speech-2.6-turbo`, `speech-2.6-hd` models Optional values: `happy`, `sad`, `angry`, `fearful`, `disgusted`, `surprised`, `calm`, `fluent`, `whisper`
• **Chinese**:
• moss\_audio\_ce44fc67-7ce3-11f0-8de5-96e35d26fb85
• moss\_audio\_aaa1346a-7ce7-11f0-8e61-2e6e3c7ee85d
• Chinese (Mandarin)\_Lyrical\_Voice
• Chinese (Mandarin)\_HK\_Flight\_Attendant
• English:
• English\_Graceful\_Lady
• English\_Insightful\_Speaker
• English\_radiant\_girl
• English\_Persuasive\_Man
• moss\_audio\_6dc281eb-713c-11f0-a447-9613c873494c
• moss\_audio\_570551b1-735c-11f0-b236-0adeeecad052
• moss\_audio\_ad5baf92-735f-11f0-8263-fe5a2fe98ec8
• English\_Lucky\_Robot
• Japanese:
• Japanese\_Whisper\_Belle
• moss\_audio\_24875c4a-7be4-11f0-9359-4e72c55db738
• moss\_audio\_7f4ee608-78ea-11f0-bb73-1e2a4cfcd245
• moss\_audio\_c1a6a3ac-7be6-11f0-8e8e-36b92fbb4f95
. This field is not returned when request fails Note: The download URL is valid for 9 hours (32400 seconds) from generation. After expiration, the file will become invalid and generated information will be lost. Please pay attention to download timing
• `0`: Success
• `1002`: Rate limit
• `1004`: Authentication failed
• `1039`: TPM rate limit triggered
• `1042`: Invalid characters exceed 10%
• `2013`: Parameter error
Options: `speech-02-hd`, `speech-02-turbo`, `speech-2.5-hd-preview`, `speech-2.5-turbo-preview`
text) and preview model (model) are provided in the request body, this parameter returns the preview audio as a URL.
voice\_id.
Enum: `png`, `webp`, `jpeg`
Enum: `png`, `webp`, `jpeg`
Accepted values: * fast\_mode is false: `360p`, `540p`, `720p`, `1080p` * fast\_mode is true: `360p`, `540p`, `720p`
Accepted values: `anime`, `3d_animation`, `clay`, `comic`, `cyberpunk`
Accepted values: `16:9`, `4:3`, `1:1`, `3:4`, `9:16`
Accepted values: * fast\_mode is false: `360p`, `540p`, `720p`, `1080p` * fast\_mode is true: `360p`, `540p`, `720p`
Accepted values: `anime`, `3d_animation`, `clay`, `comic`, `cyberpunk`
Enum: `jpeg`, `png`, `webp`
Enum: `png`, `webp`, `jpeg`
Enum: `png`, `webp`, `jpeg`
Enum: `png`, `webp`, `jpeg`
Enum: `png, webp, jpeg`
Enum: `png`, `webp`, `jpeg`
Enum: `png`, `webp`, `jpeg`
Supported values:
* `"url"`: Returns a downloadable JPEG image link. * `"b64_json"`: Returns the image data as a Base64-encoded JSON string.
Recommended aspect ratios and resolutions:
* `1024x1024` (1:1) * `864x1152` (3:4) * `1152x864` (4:3) * `1280x720` (16:9) * `720x1280` (9:16) * `832x1248` (2:3) * `1248x832` (3:2) * `1512x648` (21:9)
* `false`: No watermark. * `true`: Adds a watermark with the label "AI generated" in the bottom-right corner.
An input image must meet the following requirements: * Image format: jpeg, png * Aspect ratio (width/height): In the range \[1/3, 3] * Width and height (px): > 14 * Size: No more than 10 MB * The value of total pixels: No more than 6000×6000 * supports uploading a maximum of 10 reference images.
Enum: `TASK_STATUS_QUEUED`, `TASK_STATUS_SUCCEED`, `TASK_STATUS_FAILED`, `TASK_STATUS_PROCESSING`
1. Video Generator APIs;
2. Text-to-Image API and Image-to-Image API.
Enum: `jpeg`, `png`, `webp`
Enum: `mp4`, `gif`
Enum: `wav`
Enum: `0`, `1`, `2`
Enum: `png`, `jpeg`
Enum: `DPM++ 2M Karras`, `DPM++ SDE Karras`, `DPM++ 2M SDE Exponential`, `DPM++ 2M SDE Karras`, `Euler a`, `Euler`, `LMS`, `Heun`, `DPM2`, `DPM2 a`, `DPM++ 2S a`, `DPM++ 2M`, `DPM++ SDE`, `DPM++ 2M SDE`, `DPM++ 2M SDE Heun`, `DPM++ 2M SDE Heun Karras`, `DPM++ 2M SDE Heun Exponential`, `DPM++ 3M SDE`, `DPM++ 3M SDE Karras`, `DPM++ 3M SDE Exponential`, `DPM fast`, `DPM adaptive`, `LMS Karras`, `DPM2 Karras`, `DPM2 a Karras`, `DPM++ 2S a Karras`, `Restart`, `DDIM`, `PLMS`, `UniPC`
Enum: `Latent`, `ESRGAN_4x`, `R-ESRGAN 4x+`, `R-ESRGAN 4x+ Anime6B`
Enum: `1`, `2`
Enum: `sd_xl_refiner_1.0.safetensors`
Enum: `none`, `canny`, `depth`, `depth_leres`, `depth_leres++`, `hed`, `hed_safe`, `mediapipe_face`, `mlsd`, `normal_map`, `openpose`, `openpose_hand`, `openpose_face`, `openpose_faceonly`, `openpose_full`, `clip_vision`, `color`, `pidinet`, `pidinet_safe`, `pidinet_sketch`, `pidinet_scribble`, `scribble_xdog`, `scribble_hed`, `segmentation`, `threshold`, `depth_zoe`, `normal_bae`, `oneformer_coco`, `oneformer_ade20k`, `lineart`, `lineart_coarse`, `lineart_anime`, `lineart_standard`, `shuffle`, `tile_resample`, `invert`, `lineart_anime_denoise`, `reference_only`, `reference_adain`, `reference_adain+attn`, `inpaint`, `inpaint_only`, `inpaint_only+lama`, `tile_colorfix`, `tile_colorfix+sharp`, `depth_anything`
Enum: `0`, `1`, `2`
Enum: `0`, `1`, `2`
By default, the generated outputs will be saved to Novita AI Storage temporarily and privately.
If set to False, Novita AI will create an additional directory in the path to save the generated outputs. The default is False.
1: Explicit Nudity, Explicit Sexual Activity, Sex Toys; Hate Symbols; Non-Explicit Nudity, Obstructed Intimate Parts, Kissing on the Lips.
2: Explicit Nudity, Explicit Sexual Activity, Sex Toys; Hate Symbols; Non-Explicit Nudity, Obstructed Intimate Parts, Kissing on the Lips; Female Swimwear or Underwear, Male Swimwear or Underwear.Enum: `0, 1, 2`
Enum: `wav`, `mp3`
Enum: `TASK_STATUS_SUCCEED`, `TASK_STATUS_FAILED`
Enum: `Emily`, `James`, `Olivia`, `Michael`, `Sarah`, `John`
Enum: `en-US`, `zh-CN`, `ja-JP`
Enum: `mp4`, `gif`
Enum: `TASK_STATUS_SUCCEED`, `TASK_STATUS_FAILED`
Enum: `png`, `webp`, `jpeg`
Enum: `TASK_STATUS_SUCCEED`, `TASK_STATUS_FAILED`
By default, the generated outputs will be saved to Novita AI Storage temporarily and privately.
If set to False, Novita AI will create an additional directory in the path to save the generated outputs. The default is False.
1: Explicit Nudity, Explicit Sexual Activity, Sex Toys; Hate Symbols; Non-Explicit Nudity, Obstructed Intimate Parts, Kissing on the Lips.
2: Explicit Nudity, Explicit Sexual Activity, Sex Toys; Hate Symbols; Non-Explicit Nudity, Obstructed Intimate Parts, Kissing on the Lips; Female Swimwear or Underwear, Male Swimwear or Underwear.
Enum: `0`, `1`, `2`
Enum: `RealESRGAN_x4plus_anime_6B`, `RealESRNet_x4plus`, `4x-UltraSharp`
Enum: `mp4`, `gif`
Accepted values: `auto`, `small`, `medium`, `large`
Acceptable values: `true`, `false` When true, the system will automatically add a suitable BGM. BGM has no time limit and the system automatically adapts.
Accepted values: `16:9`, `9:16`, `1:1`
Accepted values: `auto`, `small`, `medium`, `large`
Acceptable values: `true`, `false` When true, the system will automatically add a suitable BGM. BGM has no time limit and the system automatically adapts.
Accepted values: `auto`, `small`, `medium`, `large`
Acceptable values: `true`, `false` When true, the system will automatically add a suitable BGM. BGM has no time limit and the system automatically adapts.
Accepted values: `auto`, `small`, `medium`, `large`
Acceptable values: `true`, `false` When true, the system will automatically add a suitable BGM. BGM has no time limit and the system automatically adapts.
Accepted values: `16:9`, `9:16`, `1:1`
Accepted values: `auto`, `small`, `medium`, `large`
Acceptable values: `true`, `false` When true, the system will automatically add a suitable BGM. BGM has no time limit and the system automatically adapts.
Accepted values: `auto`, `small`, `medium`, `large`
Acceptable values: `true`, `false` When true, the system will automatically add a suitable BGM. BGM has no time limit and the system automatically adapts.
Accepted values: `general`, `anime` * `general`: General style. Allows style control through prompts * `anime`: Anime style. Optimized for anime aesthetics, with better performance for anime-related prompts
Accepted values: `16:9`, `9:16`, `1:1`
Accepted values: `auto`, `small`, `medium`, `large`
Acceptable values: `true`, `false` When true, the system will automatically add a suitable BGM. BGM has no time limit and the system automatically adapts.
Options: `480P`, `720P`, `1080P`
Example: `ostris/wan22_i2v_14b_orbit_shot_lora`.
Example: `Cseti/wan2.2-14B-Kinestasis_concept-lora-v1`.
Options: `480P`, `720P`, `1080P`. Default value: `1080P`.
now you can also listen to **"ASYNC\_TASK\_RESULT"** event on your webhook endpoint. Also, you can test your integration without making actual API calls by enabling Test Mode, which will help you thoroughly test your logic.
More details can be found at: [Webhook](/api-reference/model-apis-webhook). # June 6, 2025 Product Updates Source: https://novita.ai/docs/changelog/02-06-25--06-06-25 ## **New Features** * **Recommend GPUs according to model templates** Recommend GPUs based on model templates to help developers quickly decide on GPU selection. Updated time: June 6, 2025 * **New API Error Code Reference for Model Services** To enhance the developer experience, we’ve added a detailed list of API error codes for **image, video, audio, and large language model services** in our Documentation Center. It covers common issues like request validation, format limits, and content compliance, helping you troubleshoot more efficiently. Visit the [Documentation](https://novita.ai/docs/api-reference/basic-error-code) Center to learn more. Updated time: June 6, 2025 # July 03, 2025 Notice Source: https://novita.ai/docs/changelog/03-07-25 ### **Model Decommissioning Notice** To continuously enhance platform service efficiency, focus on core product capability upgrades, and deliver more advanced, high-quality, and standardized services, Novita will decommission certain large language models (LLMs) on **July 5, 2025(UTC 0:00)**. **The deprecated models and their corresponding replacements** are listed below: | **Deprecated Models** | **Deprecation Date** | **Replacement Models** | | :------------------------------- | :------------------- | :--------------------- | | thudm/glm-z1-9b-0414 | 2025-7-5 | qwen/qwen3-4b-fp8 | | thudm/glm-4-9b-0414 | 2025-7-5 | qwen/qwen3-4b-fp8 | | thudm/glm-z1-rumination-32b-0414 | 2025-7-5 | qwen/qwen3-32b-fp8 | | thudm/glm-z1-32b-0414 | 2025-7-5 | qwen/qwen3-30b-a3b-fp8 | # Nov 3, 2025 Product Updates Source: https://novita.ai/docs/changelog/03-11-25 **Purpose** Storage-related costs for Agent Sandbox have continued to increase, and the platform can no longer maintain free storage usage.To ensure sustainable operations and more efficient resource allocation, the Agent Sandbox Storage Billing feature has been launched. **Billing Scope** * Storage space used by Sandbox Templates * Files and data retained when a Sandbox is paused **Free Period** A half-month free period is provided until November 17, 2025 (UTC). During this period, users will not be charged and can check their usage at: [https://novita.ai/sandbox-console/usage](https://novita.ai/sandbox-console/usage) **Next Steps** After the free period, storage billing will take effect. Users will be notified in advance via in-app messages or email. Updated time: Nov 3, 2025 # August 8, 2025 Notice Source: https://novita.ai/docs/changelog/04-08-25--08-08-25 ## **Model APIs Updates** ### **New Features** * **New Model Available: Qwen-Image** The **Qwen-Image** is a 20B MMDIT model for next-gen text-to-image generation. Especially strong at creating stunning graphic posters with native text. Key Highlights: * SOTA text rendering — rivals GPT-4o in English, best-in-class for Chinese * In-pixel text generation — no overlays, fully integrated * Bilingual support, diverse fonts, complex layouts Also excels at general image generation — from photorealistic to anime, impressionist to minimalist. A true creative powerhouse. Get started now: [Qwen-Image](https://novita.ai/docs/api-reference/model-apis-qwen-txt2img#qwen-image-text-to-image) Updated time: August 6, 2025 * **New Model Available: Wan 2.2 Text to Video, Wan 2.2 Image to Video** Wan2.2 is the world's first open-source MoE-architecture video generation model with cinematic control. Its Mixture-of-Experts (MoE) architecture scales model capacity without increasing computational requirements. Key Highlights: * First-of-its-kind MoE Architecture: The first model to apply an MoE framework to video generation, drastically cutting down on resource consumption. * 50% More Efficient: Saves approximately 50% of computational resources compared to models of a similar parameter scale. * Enhanced Capabilities: Achieves significant improvements in generating complex motion, character interactions, and overall aesthetic quality. Get started now: [Wan 2.2 Text to Video](https://novita.ai/docs/api-reference/model-apis-wan2.2-t2v), [Wan 2.2 Image to Video](https://novita.ai/docs/api-reference/model-apis-wan2.2-i2v) Updated time: August 8, 2025 *** ## **GPUs Updates** ### **New Features** * **Add region selection when creating an endpoint in Serverless** Novita now supports creating endpoints in selected regions. Updated time: August 5, 2025 # February 6, 2025 Source: https://novita.ai/docs/changelog/06-02-25 **Announcement: qwen/qwen-2-72b-instruct Model Decommission Notice** We plan to decommission the **qwen/qwen-2-72b-instruct** model on **February 6 2025, at 23:59 Pacific Standard Time (PST).** Please make any necessary preparations in advance. Thank you for your understanding and support! # May 9, 2025 Product Updates Source: https://novita.ai/docs/changelog/06-05-25--09-05-25 ## **Model APIs Updates** ### **New Features** * **Added filters to the Model Library for easier model selection** **Model API now features a newly launched filtering system!** Users can now filter models in the Model Library by **provider, pricing, context length, and model specifications**, enabling faster discovery and more efficient model selection and usage. Updated time: May 9, 2025 # September 6, 2024 Source: https://novita.ai/docs/changelog/06-09-24 📢 Video Merge Face API is now available on Novita AI. You can integrate them via API: [https://novita.ai/api-reference/model-apis-video-merge-face](/api-reference/model-apis-video-merge-face). # April 7, 2024 Source: https://novita.ai/docs/changelog/07-04-24 Now the LLM Chat API supports seamless integration into your product, offering compatibility with the OpenAI API standard. 🚀
More details: [LLM API Docmentation](/guides/llm-api). # April 11, 2025 Product Updates Source: https://novita.ai/docs/changelog/07-04-25--11-04-25 ## **New Features** * **Template Library Added to Console** 1. We've added a Template Library to the console, featuring templates created by both the official team and the user community. 2. In addition to deploying and saving templates, users are encouraged to **share templates** they like by copying the link **with a single click**. 3. To contribute your own templates, simply **set the visibility** to Public when creating one. Public templates will appear in the Template Library, and those that gain popularity may be featured. Updated time: April 7, 2025 * **New Model: meta-llama/llama-4-maverick-17b-128e-instruct-fp8** Novita has deployed **Llama 4 Maverick 17B Instruct (128E)** , a high-performance multimodal language model developed by Meta using a **Mixture-of-Experts (MoE)** architecture. The model dynamically activates 17 billion parameters per inference from a total of 400 billion parameters, distributed across 128 expert modules. It supports **text and image inputs in 12 languages**, and produces **text and code outputs**. Designed specifically for **vision-language tasks**, the model has been instruction-tuned to handle assistant-style interactions, image reasoning, and a wide range of general multimodal applications. Updated time: April 7, 2025 * **New Model: meta-llama/llama-4-scout-17b-16e-instruct** Novita has deployed **Llama 4 Scout 17B Instruct (16E)**, a **Mixture-of-Experts (MoE)** language model developed by Meta. The model dynamically activates 17 billion parameters per inference from a total of 109 billion parameters, utilizing 16 expert modules each time. It supports **multimodal inputs** (text and images) and generates **text or code outputs** in **12 languages**. Designed for **intelligent assistant interactions** and **visual reasoning tasks**, It also supports a **10 million-token context window**, and was trained on a **massive 40-trillion-token dataset**. Updated time: April 7, 2025 ## **Improvements** * **Improved Instance Status Visibility** Two new intermediate statuses—**“Starting”** and **“Stopping”**—have been introduced to provide **more granular visibility** during instance state transitions. Updated time: April 10, 2025 # July 11, 2025 Product Updates Source: https://novita.ai/docs/changelog/07-07-25--11-07-25 ## **Model APIs Updates** ### **New Features** * **MiniMax Speech-02 and MiniMax Voice Cloning Now Available on Novita** Novita now supports **MiniMax Speech-02-HD** and **MiniMax Speech-02-Turbo** audio generation models, along with the **MiniMax Voice Cloning** model. **MiniMax Speech-02** features powerful generalization capabilities, effortlessly handling speech across 32 languages, various accents, and a wide range of emotional tones. Since its release in May, it has consistently ranked No.1 globally on the Artificial Analysis leaderboard. **MiniMax Voice Cloning** is designed for use cases such as IP voice replication and timbre cloning, offering fast and accurate voice reproduction. It supports both mono and stereo voice cloning, enabling quick replication of a target voice from an audio sample. Get started now: [**MiniMax Speech-02-HD**](https://novita.ai/docs/api-reference/model-apis-minimax-speech-02-hd)\[[**********; MiniMax Speech-02-Turbo; MiniMax Voice Cloning.**********](https://novita.ai/docs/api-reference/model-apis-minimax-voice-cloning)]\([https://novita.ai/docs/api-reference/model-apis-minimax-speech-02-turbo](https://novita.ai/docs/api-reference/model-apis-minimax-speech-02-turbo)) Updated time: July 10, 2025 * **GLM-4.1V-9B-Thinking Now Available on Novita** As a lightweight model with 9 billion parameters, **GLM-4.1V-9B-Thinking** delivers impressive performance—matching or even surpassing the 72B-parameter Qwen-2.5-VL-72B in 18 out of 28 authoritative benchmark evaluations. It excels in image-text understanding, scientific and mathematical reasoning, and video comprehension tasks. The model also supports ultra-high-resolution images up to 4K and accepts inputs with arbitrary aspect ratios. Get started now: [**GLM-4.1V-9B-Thinking**](https://novita.ai/models/llm/thudm-glm-4.1v-9b-thinking)**.** Updated time: July 10, 2025 * **Kimi-K2-Instruct Now Available on Novita** **Kimi K2** is a state-of-the-art Mixture-of-Experts (MoE) language model featuring 32 billion active parameters and 1 trillion total parameters. Trained with the Muon optimizer, Kimi K2 delivers exceptional performance in frontier knowledge tasks, reasoning, and coding. It is also carefully optimized for agent capabilities—designed specifically for tool use, advanced reasoning, and autonomous problem solving. Get started now: [**Kimi-K2-Instruct**](https://novita.ai/models/llm/thudm-glm-4.1v-9b-thinking) Updated time: July 11, 2025 *** ## **GPUs Updates** ### **New Features** * **Instance migration supports migrating system disks** Supports instance migration and also supports migrating the system disk. After migration, instance information (such as instance ID, egress IP, etc.) will not change. After migration, the mirroring will be stored in the Novita official mirroring repository. If the system disk is migrated, the system disk data will be packaged into a mirroring and migrated together to the new resource. The new mirroring contains the system disk data and is ready to use immediately. Updated time: July 7, 2025 * **GPU Instance metrics** Supports real-time monitoring of the following data: * CPU Utilization * Memory Utilization * GPU Utilization * Video Memory Utilization * System Disk Utilization Updated time: July 9, 2025 # December 7, 2024 Source: https://novita.ai/docs/changelog/07-12-24 Llama 3.3 70B is Here! We're thrilled to announce the launch of Llama 3.3 70B, a model optimized for multilingual dialogue. It delivers exceptional performance, surpassing many open-source and proprietary chat models in industry benchmarks. Pricing: **\$0.39 per 1M tokens** (input & output). # September 12, 2025 Product Updates Source: https://novita.ai/docs/changelog/08-09-25--12-09-25 ## **Model APIs Updates** ### **New Features** * **Wan2.2 I2V and Wan2.2 T2V supported LoRA** Wan2.2 I2V and Wan2.2 T2V supported LoRA. 1080P is not supported. List of LoRAs to apply (max 3). Updated time: September 9, 2025 # January 9, 2025 Source: https://novita.ai/docs/changelog/09-01-25 For improved security, **API Keys** will no longer be visible in plaintext after creation. Please make sure to save your key when it's generated, as it cannot be retrieved afterward. # May 9, 2024 Source: https://novita.ai/docs/changelog/09-05-24 Now Novita AI has released the Text-To-Speech API. Text-To-Speech API generates realistic speech in various languages
👉 [Text-To-Speech API Docmentation](/api-reference/model-apis-txt2speech). # March 14, 2025 Product Updates Source: https://novita.ai/docs/changelog/10-03-25--14-03-25 ## **New Features** * **Embedding Model-BAAI: BGE-M3 is Added to Model Library** Novita supports the `baai/bge-m3` text embedding model. This model is capable of simultaneously achieving three core functionalities: dense retrieval, multi-vector retrieval, and sparse retrieval. Designed to be compatible with over 100 languages, it supports a variety of input formats, ranging from short sentences to long texts of up to 8,192 tokens. For detailed usage instructions, please refer to the API documentation: [Model-BAAI: BGE-M3](https://novita.ai/docs/api-reference/model-apis-llm-create-embeddings) Updated time: March 13, 2025 * **Login Security: Limit on Incorrect Password Attempts** To enhance account security, a login attempt limit has been implemented: Maximum 5 incorrect passwords within 1 hour. Updated time: March 11, 2025 ## **Improvements** * **Pricing Update for Wan 2.1 T2V Video Model** Optimized pricing strategy for video generation models: The pricing for all resolutions is now standardized at \$0.08 per second. Updated time: March 10, 2025 # April 11, 2024 Source: https://novita.ai/docs/changelog/11-04-24 We've recently added more hot LLM models for you, now the supported models include: * lzlv\_70b ([https://huggingface.co/lizpreciatior/lzlv\\\_70b\\\_fp16\\\_hf](https://huggingface.co/lizpreciatior/lzlv\\_70b\\_fp16\\_hf)) * Nous-Hermes-2-Mixtral-8x7B-DPO ([https://huggingface.co/NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO](https://huggingface.co/NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO)) * gryphe/mythomax-l2-13b ([https://huggingface.co/Gryphe/MythoMax-L2-13b](https://huggingface.co/Gryphe/MythoMax-L2-13b)) * nousresearch/nous-hermes-llama2-13b ([https://huggingface.co/NousResearch/Nous-Hermes-Llama2-13b](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-13b)) # August 15, 2025 Product Updates Source: https://novita.ai/docs/changelog/11-08-25--15-08-25 ## **Model APIs Updates** ### **New Features** * **Batch Inference API for Large Language Models** Our Batch Inference API enables asynchronous processing of high-volume inference requests while maintaining full compatibility with the OpenAI API standard. This cost-efficient solution is ideal when immediate results aren't required, offering significantly higher rate limits than online calls while guaranteeing delivery within a reasonable 48-hour window. Perfect for: • Evaluation and data analysis at scale • Large dataset classification • Offline document summarization generation **Pricing:** Available at 50% off standard model inference API rates. Updated time: August 13, 2025 * **New Model Available: Vidu Q1** **Vidu Q1** is a multimodal video generation model globally launched by Shengshu Technology in April 2025. It features direct output capability at 1080P resolution and AI sound synchronization functionality. The model achieves state-of-the-art (SOTA) performance in video quality and semantic consistency metrics on VBench-1.0, as well as in commonsense reasoning and physical understanding dimensions on VBench-2.0. Additionally, it ranks first in both anime and realistic style categories on the SuperCLUE text-to-video leaderboard in China. Discover the power of Vidu Q1 on Novita by exploring its [text-to-video](https://novita.ai/docs/api-reference/model-apis-vidu-q1-txt2video), [image-to-video](https://novita.ai/docs/api-reference/model-apis-vidu-q1-img2video), [start-end-to-video](https://novita.ai/docs/api-reference/model-apis-vidu-q1-startend2video), and [reference-to-video](https://novita.ai/docs/api-reference/model-apis-vidu-q1-reference2video) generation capabilities—and bring your creative ideas to life. Updated time: August 14, 2025 * **New Model Available: Vidu 2.0** **Vidu 2.0** is a multimodal video generation model globally launched by Shengshu Technology in January 2025. It can generate a single 4-second video clip at 512P resolution within 10 seconds, delivering more consistent styles and themes, smoother transitions between the first and last frames, and producing four materials in one go—doubling efficiency. Discover the efficiency of Vidu 2.0 on this platform by exploring its [image-to-video](https://novita.ai/docs/api-reference/model-apis-vidu-2.0-img2video), [start-end-to-video](https://novita.ai/docs/api-reference/model-apis-vidu-2.0-startend2video), and [reference-to-video](https://novita.ai/docs/api-reference/model-apis-vidu-2.0-reference2video) generation capabilities—and see how effortless video creation can be. Updated time: August 14, 2025 # November 11, 2024 Source: https://novita.ai/docs/changelog/11-11-24 Building on our last update, where we brought you unbeatable pricing on Llama 3.1 8B, our team has kept working hard to bring even more savings across our platform! 📉 New Pricing for LLama 3.1 70B Instruct * Input: **\$0.34/Mtokens** (↓33.3% from \$0.51) * Output: **\$0.39/Mtokens** (↓47.2% from \$0.74) These new rates are live on both Novita AI and OpenRouter, where we've secured the #1 spot for competitive pricing. # May 16, 2025 Product Updates Source: https://novita.ai/docs/changelog/12-05-25--16-05-25 ## **Model APIs Updates** ### **Improvements** * **Hunyuan Video Fast Now 67% Cheaper** We’ve significantly reduced the pricing for the Hunyuan Video Fast model — **now only one-third of the original price**. Enjoy high-quality video generation at a much lower cost. Updated time: May 13, 2025 *** ## **GPUs Updates** ### **New Features** * **Compute Capacity Expanded to Cover All Major Continents** Our global compute footprint continues to grow. Leveraging a distributed resource pool, we now offer GPU services across all major continents—empowering customers worldwide. **New Regions:** 1. **Middle East:** United Arab Emirates (AS-AE-01) 2. **Europe:** United Kingdom (EU-GB-01) 3. **South Asia:** India (AS-IN-01) 4. **South America:** Brazil (SA-BR-01) Updated time: May 13, 2025 ### **Improvements** * **Explore & Instances Interface Enhancements** We’ve refined both the Explore and Instances pages to surface more information and streamline interactions: 1. Display architecture names for H100, H200, and A100 GPU card types 2. On the Explore page, display the maximum GPUs supported per server as "\[card count] max" and streamline the per-instance GPU selector. 3. On the Instances page, display the number of GPUs configured for each instance Updated time: May 15, 2025 # February 14, 2025 Source: https://novita.ai/docs/changelog/14-02-25 We've introduced a referral program for developers. Refer a friend to Novita, and both of you will receive \$20 in DeepSeek API credits, with a total of up to \$500 in credits available. For more details, please refer to [Referral Program](https://novita.ai/referral). # March 14, 2024 Source: https://novita.ai/docs/changelog/14-03-24 We've released novita.ai Javascript SDK version 1.0.35, available at [https://www.npmjs.com/package/novita-sdk/v/1.0.35](https://www.npmjs.com/package/novita-sdk/v/1.0.35). This version introduces the img2ImgV3 and txt2ImgV3 methods for calling the API v3 txt2img and img2img,
and you can find examples here: [https://github.com/novitalabs/javascript-sdk/blob/main/examples](https://github.com/novitalabs/javascript-sdk/blob/main/examples). # April 18, 2025 Product Updates Source: https://novita.ai/docs/changelog/14-04-25--18-04-25 ## **New Features** * **Sign Up with Google, GitHub, or Hugging Face Now Supported** Users can now register quickly and securely using their existing **Google**, **GitHub**, or **Hugging Face** accounts—no need to create a new password. Updated time: April 17, 2025 * **Rate Limit Display Added for LLM Usage in Console** The console now displays **rate limit information** for large language models (LLMs), allowing users to easily view the default request limits for each model and better plan their usage. In addition, a new rate limit increase request entry has been added, enabling users to submit requests for higher limits based on specific business needs. Updated time: April 17, 2025 * **New Video Model Added: Wan 2.1 I2V** The **Wan 2.1 I2V (Image-to-Video)** model is now available on the platform. It enables the generation of continuous video sequences from a single input image. This model delivers strong performance in **motion consistency**, **visual stability**, and **detail fidelity**, making it well-suited for **creative content generation**, **virtual character animation**, and related use cases. Updated time: April 16, 2025 * **Pricing Update for Video Model Wan 2.1 T2V** We’ve introduced a **resolution-based pricing model** for the **Wan 2.1 T2V (Text-to-Video)** model.\ Users are now charged based on the output video resolution, enabling more flexible and cost-effective usage based on quality requirements. Updated time: April 16, 2025 * **Function Calling Now Supported on meta-llama/llama-4-maverick-17b-128e-instruct-fp8** The model `meta-llama/llama-4-maverick-17b-128e-instruct-fp8` supports the function call feature. For usage instructions, please refer to the [documentation](https://novita.ai/docs/guides/llm-function-calling). Updated time: April 14, 2025 * **Function Calling Now Supported on meta-llama/llama-4-scout-17b-16e-instruct** The model `meta-llama/llama-4-scout-17b-16e-instruct` supports the function call feature. For usage instructions, please refer to the [documentation](https://novita.ai/docs/guides/llm-function-calling). Updated time: April 14, 2025 # May 14, 2025 Notice Source: https://novita.ai/docs/changelog/14-05-25 ### **Model Decommissioning Notice** To continuously enhance platform service efficiency, focus on core product capability upgrades, and deliver more advanced, high-quality, and standardized services, Novita will decommission certain large language models (LLMs) on **May 28, 2025(UTC 00:00)** . **The deprecated models and their corresponding replacements** are listed below: | **Deprecated Models** | **Deprecation Date** | **Replacement Models** | | :--------------------------------------- | :------------------- | :-------------------------------- | | jondurbin/airoboros-l2-70b | 2025-5-28 | sao10k/l3-70b-euryale-v2.1 | | qwen/qwen3-14b-fp8 | 2025-5-28 | qwen/qwen3-32b-fp8 | | qwen/qwen3-0.6b-fp8 | 2025-5-28 | qwen/qwen3-32b-fp8 | | qwen/qwen3-1.7b-fp8 | 2025-5-28 | qwen/qwen3-32b-fp8 | | meta-llama/llama-3.2-11b-vision-instruct | 2025-5-28 | qwen/qwen2.5-vl-72b-instruct | | google/gemma-2-9b-it | 2025-5-28 | google/gemma-3-27b-it | | meta-llama/llama-3.1-70b-instruct | 2025-5-28 | meta-llama/llama-3.3-70b-instruct | | qwen/qwq-32b | 2025-5-28 | meta-llama/llama-3.3-70b-instruct | # July 18, 2025 Product Updates Source: https://novita.ai/docs/changelog/14-07-25--18-07-25 ## **GPUs Updates** ### **New Features** * **GPU Instance Spot Pricing** Advantages of Functional Features: 1. Greatest Advantage: **Affordable!! The price is generally 50% off the on-demand price**, making it highly price-competitive in the market. For example, the current price of 4090 is \$0.35, while the spot 4090 price is as low as \$0.18, and larger usage volumes may result in greater discounts. 2. Why is it so cheap? Spot instances are essentially designed to handle idle inventory, so **they may be reclaimed at any time due to resource shortages** (instance termination). 3. How can we ensure the stability of our customers' business as much as possible while keeping costs low? Our spot billing instances will adopt the following strategies: * After creation, it enjoys **a 1-hour grace period**, during which the instance is guaranteed not to be reclaimed * Before resource recycling, a notice will be sent **1 hour in advance, and another notice will be sent 5 minutes before instance recycling**. When customers receive the notice, they can back up their data, while we give customers more time to back up their data! Updated time: July 18, 2025 *** ## **Agent SandBox Updates** ### **New Features** * **The product page for Agent Sandbox is now live** This page provides a concise overview of Agent Sandbox's core features, use cases, pricing model, and helpful documentation, enabling users to quickly get started and explore its powerful capabilities. **Agent Sandbox** is a high-performance cloud runtime purpose-built for **AI Agent workloads**, designed to support the full lifecycle of autonomous agents. Key features include: * **Millisecond-Level Startup** * **Secure Isolation** * **High Concurrency** Explore the [Agent Sandbox](https://novita.ai/sandbox) product page to discover more details! Updated time: July 18, 2025 # September 19, 2025 Product Updates Source: https://novita.ai/docs/changelog/15-09-25--19-09-25 ## **Model APIs Updates** ### **New Features** * **Seedream-4.0 Now Available** Seedream is a state-of-the-art (SOTA) multimodal image creation model based on a cutting-edge architecture. It breaks the creative boundaries of traditional text-to-image models by natively supporting text, single-image, and multi-image inputs. Users can freely combine text and images to enable diverse creative approaches within the same model, including multi-image fusion creation with subject consistency, image editing, and batch image generation. This makes image creation more free and controllable. Charges are based on the actual number of images generated; no fees will be incurred for failed generation. Updated time: September 19, 2025 *** ## **Agent SandBox Updates** ### **New Features** * **Metrics History Available for Paused Sandboxes** When a Sandbox is in Paused state, you can now access metrics data from up to 72 hours before the last pause time. This enhancement allows developers to review and analyze recent activity even after a Sandbox has been paused. Updated time: September 18, 2025 # January 16, 2025 Source: https://novita.ai/docs/changelog/16-01-25 We're excited to bring DeepSeek V3, a state-of-the-art MoE language model, to Novita AI! Why Choose DeepSeek V3? * A robust 671B total parameter model, outperforming open-source peers and rivaling top-tier closed-source models. * Performs exceptionally well for coding tasks, making it a great companion for developers. * Handle complex inputs with 16K context size and 16K max output. * Just \$0.89/M tokens, delivering exceptional value without compromising performance. How to Access DeepSeek-V3: * Explore it through our Playground: [https://novita.ai/models/llm/deepseek-deepseek\_v3](https://novita.ai/models/llm/deepseek-deepseek_v3); * Use it seamlessly via OpenRouter: [https://openrouter.ai/deepseek/deepseek-chat](https://openrouter.ai/deepseek/deepseek-chat). Experience the future of AI—try DeepSeek V3 today! 🚀 # June 20, 2025 Product Updates Source: https://novita.ai/docs/changelog/16-06-25--20-06-25 ## **Model APIs Updates** ### **New Features** * **MiniMax-Hailuo-02 Now Available on Novita** Novita now supports MiniMax’s latest high-performance video generation model — **Hailuo 02**. Hailuo 02 delivers industry-leading instruction-following capabilities and excels at generating videos in complex physical scenarios, including high-intensity stunt simulations. It supports ultra-high-definition video generation, with: * Up to **6 seconds** at **768P or 1080P** * Up to **10 seconds** at **768P** Explore more creative possibilities with one of the most advanced video models available. Get started now: [**MinMax-Hailuo-02**](https://novita.ai/models/image#minimax-hailuo-02) Updated time: June 18, 2025 *** ## **GPUs Updates** ### **New Features** * **Instances are allowed to upgrade in any state** Updated time: June 17, 2025 # Weekly Product Updates - March 21, 2025 Source: https://novita.ai/docs/changelog/17-03-25--21-03-25 ## **New Features** * **LLM Full Model Upgrade: Function Calling & Structured Outputs Available Now** Novita now supports Function Calling and Structured Outputs for all capability-supported models on the platform. Developers can now filter models using these tags for more precise API integration and structured data generation that aligned with custom schemas. Refer to the API documentation for details: [**Function calling**](https://novita.ai/docs/guides/llm-function-calling) & [**Structured Outputs**](https://novita.ai/docs/guides/llm-structured-outputs) Updated time: March 21, 2025 * **Image Dedicated Endpoint now supports billing queries** Billing for Image Dedicated Endpoints now includes monthly query support. Updated time: March 21, 2025 * **The annual and monthly subscription functionality of GPU Instance is now officially released** 1. **Flexible Billing Model:** Users can freely choose between on-demand and monthly subscription when launching instances. 2. **Enhanced Resource Guarantee:** During the monthly subscription period, even if an instance is shut down, it remains reserved for the user, significantly enhancing the user experience. 3. **Seamless Service Transition:** Supports converting on-demand instances to annual or monthly subscriptions, as well as renewals during the subscription period. 4. **Monthly Subscription Discounts:** Monthly subscriptions start at a 10% discount over pay-as-you-go, with additional saving for longer terms. Updated time: March 19, 2025 * **Team members' nicknames are now displayed in the task lists for Instance, Serverless GPUs, and more** When the account is in Team mode, the GPUs Console now displays member nicknames. Updated time: March 18, 2025 # August 22, 2025 Product Updates Source: https://novita.ai/docs/changelog/18-08-25--22-08-25 ## **Model APIs Updates** ### **New Features** * **New Model Available: MiniMax Speech-2.5** Novita now features MiniMax's next-generation Text-to-Speech models: **Speech-2.5-HD** and **Speech-2.5-Rurbo**. Speech 2.5 represents a major breakthrough in voice generation, delivering significant advancements in multi-language expressiveness, voice cloning fidelity, and language coverage. Speech-2.5-HD (High-Fidelity):The latest HD model, offering exceptional prosody and remarkable voice cloning accuracy.Specifically designed for high-quality audio applications, providing studio-grade, high-fidelity sound. Speech-2.5-Turbo:The latest Turbo model, engineered for extreme responsiveness and ultra-low latency.Ideal for high-concurrency, real-time scenarios and includes support for 40 languages. Get started now: [MiniMax Speech-2.5-hd-preview Text to Speech](https://novita.ai/docs/api-reference/model-apis-minimax-speech-2.5-hd-preview), [MiniMax Speech-2.5-turbo-preview Text to Speech](https://novita.ai/docs/api-reference/model-apis-minimax-speech-2.5-turbo-preview) Updated time: August 21, 2025 * **New Model Available: Qwen-Image-Edit** Built upon the powerful 20B Qwen-Image model, Qwen-Image-Edit extends its unique text rendering capabilities into the image editing domain, enabling highly precise in-image text manipulation. Key Highlights: * **Comprehensive Editing:** Supports a full spectrum of edits, from precise local adjustments (adding, removing, or modifying elements while keeping the rest of the image intact) to high-level creative edits (like IP creation, object rotation, and style transfer). * **Precise Text Editing:** Directly add, delete, or modify both Chinese and English text within an image while perfectly preserving the original font, size, and style. * **SOTA Performance:** Achieves state-of-the-art (SOTA) results on multiple public benchmarks, establishing it as a powerful foundational model for image editing tasks. Get started now:[**Qwen-Image-Edit**](https://novita.ai/docs/api-reference/model-apis-qwen-image-edit) Updated time: August 21, 2025 * **New Model Available: Pixverse V4.5 Text to Video, Pixverse V4.5 Image to Video** Pixverse 4.5 is the latest AI video model released by Pixverse AI. It introduces major upgrades in cinematic control, multi-image fusion, and complex motion handling. Key Highlights: * **20+ Cinematic Camera Controls:** Access over 20 camera controls like panning and zooming, all precisely guided by text prompts for professional-grade results. * **Multi-Image Reference & Fusion:** Seamlessly fuse elements from multiple reference images with the new Fusion mode for richer, stylistically consistent videos. * **Optimized Complex Motion:** Superior handling of complex actions like high-speed movement and group interactions, delivering smoother and more natural animations. Get started now:[ Pixverse V4.5 Text to Video](https://novita.ai/docs/api-reference/model-apis-pixverse-v4.5-t2v), [PixVerse V4.5 Image to Video](https://novita.ai/docs/api-reference/model-apis-pixverse-v4.5-i2v) Updated time: August 21, 2025 *** ## **Platform Service Updates** ### **New Features** * **Budget Management Now Live for Teams** Our new Budget Management feature is now available, giving you precise control over your team's API spending. This system is designed to help you efficiently manage member-level spending limits, with key capabilities including: * **Member-Level Budget Control:** Assign unique spending caps to individual team members. * **Smart Enforcement & Alerts:** Automatically prevent cost overruns and receive timely notifications. * **Flexible Budget Types:** Implement one-time or recurring (daily, weekly, or monthly) budgets. * **Real-Time Adjustments:** Instantly modify budget allocations as your needs change. To get started, visit [Billing > Budgets](https://novita.ai/billing/budgets) in your console. For more details, please see our [help documentation](https://novita.ai/docs/guides/budgets). **Updated:** August 20, 2025 # January 19, 2026 Product Updates Source: https://novita.ai/docs/changelog/19-01-26 ## **GPUs Updates** ### **New Features** * **Added Automatic Instance Migration for newly GPU instance** The GPU **Automatic Instance Migration** entry has been relocated to the **Instance Management** page, and now supports configuring migration policies at the **instance** level. To ensure business continuity, **Automatic Instance Migration** will be enabled by default for **newly created resources starting at 2026**. You can manage migration settings via `More Operations`→ `Automatic Instance Migration` or configure migration attributes through the **OpenAPI**. Updated time: Jan 19, 2026 # April 19, 2024 Source: https://novita.ai/docs/changelog/19-04-24 🔥 Llama 3 models are now available on novita.ai. You can try them out in the playground: * meta-llama/llama-3-8b-instruct: [https://novita.ai/model-api/llm-api/playground#meta-llama-llama-3-8b-instruct](https://novita.ai/llm-api/playground#meta-llama-llama-3-8b-instruct) * meta-llama/llama-3-70b-instruct: [https://novita.ai/model-api/llm-api/playground#meta-llama-llama-3-70b-instruct](https://novita.ai/llm-api/playground#meta-llama-llama-3-70b-instruct) # May 23, 2025 Product Updates Source: https://novita.ai/docs/changelog/19-05-25--23-05-25 ## **Model APIs Updates** ### **New Features** * **Kling V1.6: Now Supports 10-Second Video Generation** Kling V1.6 now supports generating videos up to 10 seconds long, enabling more complete and coherent visual storytelling. Ideal for short narratives, ad snippets, and dynamic showcases. Get started now: [Novita AI KLING V1.6 Text to Video API | Generate videos from text](https://novita.ai/docs/api-reference/model-apis-kling-v1.6-t2v) and [Novita AI KLING V1.6 Image to Video API | Generate videos from images](https://novita.ai/docs/api-reference/model-apis-kling-v1.6-i2v). Updated time: May 21, 2025 *** ## **GPUs Updates** ### **New Features** * **New GPU Type L40S Now Available** The **L40S** is a **high-performance data center GPU** from NVIDIA, built on the Ada Lovelace architecture. It is optimized for **multi-tasking scenarios** such as AI inference, large model training, graphics rendering, and virtualization. The L40S is well-suited for **inference and training of medium to large language models** like LLaMA, GPT, and BERT, as well as for **Stable Diffusion and image/video generation**. It supports concurrent operation of multiple hardware engines and offers excellent **multi-user VM deployment** and **GPU virtualization capabilities**. Get started now: [Novita AI](https://novita.ai/gpus-console). Updated time: May 23, 2025 # July 19, 2024 Source: https://novita.ai/docs/changelog/19-07-24 Novita AI has released **"GPU Instance"**, which provides cost-effective, accessible high-performance computing with easy GPU access, ideal for complex AI tasks. Users benefit from on-demand rental, optimizing efficiency and cost-effectiveness.
For more details, please visit: [https://novita.ai/gpu-instance](https://novita.ai/gpu-instance). # October 19, 2024 Source: https://novita.ai/docs/changelog/19-10-24 We've recently deployed some new LLM models. You can try them out in the playground: * **qwen/qwen-2-72b-instruct**: [https://novita.ai/llm-api/playground#qwen-qwen-2-72b-instruct](https://novita.ai/llm-api/playground#qwen-qwen-2-72b-instruct) * **qwen/qwen-2.5-72b-instruct**: [https://novita.ai/llm-api/playground#qwen-qwen-2.5-72b-instruct](https://novita.ai/llm-api/playground#qwen-qwen-2.5-72b-instruct) * **qwen/qwen-2-7b-instruct**: [https://novita.ai/llm-api/playground#qwen-qwen-2-7b-instruct](https://novita.ai/llm-api/playground#qwen-qwen-2-7b-instruct) * **sao10k/l31-70b-euryale-v2.2**: [https://novita.ai/llm-api/playground#sao10k-l31-70b-euryale-v2.2](https://novita.ai/llm-api/playground#sao10k-l31-70b-euryale-v2.2) Or integrate with API: [LLM API Documentation](/guides/llm-api). # December 19, 2024 Source: https://novita.ai/docs/changelog/19-12-24 The **Team** feature is now available. **With Team, you can:** * Convert your personal account into a team account or join an existing team to collaborate with others; * Invite other users to join your team, and share the team's resources with them; * Manage your members' permissions to the team's resources by assigning different roles; * Analyze usage data for each team member. For more information, please visit: [Team Documentation](/guides/team). # March 20, 2024 Source: https://novita.ai/docs/changelog/20-03-24 We've released novita.ai Python SDK version 0.5.0 🐍 This version introduces the txt2img\_v3 and img2img\_v3 methods for calling the API v3 txt2img and img2img,
and you can find examples here: [https://github.com/novitalabs/python-sdk/tree/main/examples](https://github.com/novitalabs/python-sdk/tree/main/examples). # April 25, 2025 Product Updates Source: https://novita.ai/docs/changelog/21-04-25--25-04-25 ## **Model APIs Updates** ### **New Features** * **New Models: The GLM-4-0414 Series and the Qwen2.5-7B-Instruct Model** As early as March this year, Zhipu AI had announced plans to open source new GLM models in mid-April, and true to their word, the GLM-4-0414 series was launched on schedule. The newly open-sourced models cover three categories: **foundation, reasoning, and reflection**. According to official data, some models in this series can rival OpenAI’s GPT-4o and DeepSeek’s V3/R1 series in performance. As a one-stop AIGC cloud service platform, Novita promptly launched the above models along with a limited-time free version of the Qwen2.5-7B-Instruct model. Novita also carried out full-chain optimization on the models, significantly reducing inference latency while maintaining the original model accuracy. Developers can efficiently access and utilize these models through API interfaces. 1. [thudm/glm-4-9b-0414](https://novita.ai/models/llm/thudm-glm-4-9b-0414) 2. [thudm/glm-z1-9b-0414](https://novita.ai/models/llm/thudm-glm-z1-9b-0414) 3. [thudm/glm-z1-32b-0414](https://novita.ai/models/llm/thudm-glm-z1-32b-0414) 4. [thudm/glm-4-32b-0414](https://novita.ai/models/llm/thudm-glm-4-32b-0414) 5. [thudm/glm-z1-rumination-32b-0414](https://novita.ai/models/llm/thudm-glm-z1-rumination-32b-0414) Updated time: April 25, 2025 * **New Product: LLM Dedicated Endpoint** Novita introduces LLM Dedicated Endpoint – a fully managed, high-performance AI deployment solution. Designed for scaling teams and growing businesses, it delivers reliable, high-performance AI infrastructure without the complexity of self-management.\_For details, see: [Dedicated Endpoint](https://novita.ai/dedicated-endpoint) Updated time: April 22, 2025 * **Novita Referral Vouchers Now Usable Across All LLM Models** Novita's referral program vouchers can be used with all LLM models, subject to applicable rate limits. For details, see: [llm-rate-limits](https://novita.ai/docs/guides/llm-rate-limits) Updated time: April 21, 2025 *** ## **GPUs Updates** ### **Improvements** * **Optimized Storage Architecture for GPU Container Instances** To address the overlap between system disks and local disks, we have updated the instance storage policy. Effective immediately, the option to configure **a local disk for new instances has been discontinued**. Instances and templates that have already been created with local disk settings will remain unaffected. Following these changes, instances now support **system disk expansion**. For instances that have existing local disks, users can now unmount or resize the local disk space. When local storage is unmounted, all associated data will be automatically migrated to the system disk to ensure data continuity.\ For more details, please refer to the [Instance Management](https://novita.ai/gpus-console/instances) page. Updated time: April 24, 2025 # July 25, 2025 Product Updates Source: https://novita.ai/docs/changelog/21-07-25--25-07-25 ## **Model APIs Updates** ### **New Features** * **New Model Available: Qwen3 Coder 480B A35B Instruct** Novita now supports **Qwen3 Coder 480B A35B Instruct**, the cutting-edge open-source coding model from the Qwen team. This model delivers performance on par with Claude Sonnet for agentic coding, browser automation, and other core development tasks, enabling highly efficient code generation and the handling of complex problems. Get started now: [**Qwen3 Coder 480B A35B Instruct**](https://novita.ai/models/llm/qwen-qwen3-coder-480b-a35b-instruct) Updated time: July 23, 2025 * **New Model Available: Qwen3 235B A22b Thinking 2507** Novita now supports Qwen3 235B A22B Instruct 2507, a powerful Mixture-of-Experts (MoE) model designed for ultimate performance and efficiency. By activating just 22B of its 235B parameters during inference, it dynamically handles diverse tasks with incredible speed. With its native 262k-token context window, the model excels at demanding use cases, including analyzing long documents, comprehending full code repositories, and performing complex multi-paragraph reasoning. Get started now: [**Qwen3 235B A22b Thinking 2507**](https://novita.ai/models/llm/qwen-qwen3-235b-a22b-thinking-2507) Updated time: July 23, 2025 *** ## **Platform Service Updates** ### **Improvements** * **Pricing Page Optimization** Non-MoE Models are now systematically sorted by vendor/manufacturer for clear differentiation Dedicated MoE (Mixture of Experts) category added and placed at the end of the page for quick access Updated time: July 22, 2025 # January 22, 2025 Source: https://novita.ai/docs/changelog/22-01-25 Introducing the brand-new Novita AI, including our new brand logo and site style.
visit [https://novita.ai/model-api/playground#img2img](https://novita.ai/model-api/playground#img2img) to have a try. And txt2img Playground allows you to share your parameters with a share-link.
Here is my share for you: [https://novita.ai/model-api/playground?share=2930971716886178](https://novita.ai/model-api/playground?share=2930971716886178). # August 1, 2025 Notice Source: https://novita.ai/docs/changelog/28-07-25--01-08-25 ## **Model APIs Updates** ### **New Features** * **New Model Available: Seedream 3.0 Text to Image** The **Seedream 3.0 Text to Image** model is a new generation text-to-image model built on ByteDance’s large visual model. It offers accurate understanding of textual input and the ability to generate images with high text-image alignment and aesthetic quality. The model supports direct output at resolutions below 2048, with significantly improved performance across various scenarios. Get started now: [**Seedream 3.0 Text to Image**](https://novita.ai/docs/api-reference/model-apis-seedream-3-0-t2i) Updated time: August 1, 2025 # February 29, 2024 Source: https://novita.ai/docs/changelog/29-02-24 We've launched some new features lately, here are the details: 1. You can now check if the civitai models exist on novita.ai. Simply copy the civitai model page URL and paste it into the search box at [https://novita.ai/model-api/model](https://novita.ai/model-api/model). 2. SDXL Lightning LoRAs are now supported in the API /v3/async/txt2img. Available models include sdxl\_lightning\_2step\_lora, sdxl\_lightning\_4step\_lora, and sdxl\_lightning\_8step\_lora. *** [View More](https://discord.gg/QwpfJqcxCX) # April 29, 2024 Source: https://novita.ai/docs/changelog/29-04-24 Now you can chat with your favorite character by using the novita.ai LLM features.
Play it now 👉 [https://novita.ai//model-api/llm-api/playground](https://novita.ai/llm-api/playground) # July 4, 2025 Product Updates Source: https://novita.ai/docs/changelog/30-06-25-04-07-2025 ## **Model APIs Updates** ### **New Features** * **Novita LLMs Dedicated Endpoints** Fully Managed Deployment for Your Custom AI Models High cost performance * Deploy available models and LoRA adapters on customizable GPU endpoints with per-hour billing. * Lower prices, No hard rate limits, Cheaper under high utilization, Custom base models and multiple LoRAs, Predictable performance unaffected. Learn more about [Pricing](https://novita.ai/pricing) and read this [Blog](https://blogs.novita.ai/deploy-custom-base-model-with-llm-dedicated-endpoint/) to learn how to deploy a custom base model with Novita LLMs Dedicated Endpoints. Updated time: July 4, 2025 *** ## **Agent SandBox Updates** ### **New Features** * **Agent Sandbox Is Now Live** **Agent Sandbox** is a high-performance cloud runtime purpose-built for **AI Agent workloads**, designed to support the full lifecycle of autonomous agents. Key capabilities include: * Millisecond-level startup, optimized for high-concurrency tasks * **Multi-language support**, including Python, JavaScript, and C++ * Enables **code execution, web access, and system-level interactions** * **Per-second billing** for vCPU and memory — flexible, cost-efficient usage As the **infrastructure foundation for the runtime era of AI Agents**, Agent Sandbox empowers the next generation of generative agents with fast, scalable, and reliable execution capabilities. Visit the [**Documentation**](https://novita.ai/docs/guides/sandbox-overview) Center to learn more. Updated time: June 30, 2025 *** ## **Platform Service Updates** ### **Improvements** * **In-App Messaging Now Live for Critical Notifications** Our new **in-app messaging system** is now available, allowing you to receive important platform-level updates in one place, including: * **GPU failures, service interruptions, and operational alerts** * **Account-related notices**, such as coupon delivery or login activity * **Promotional events and platform campaigns** All messages can be viewed in your **Message Center** in the console. Stay informed, effortlessly. Updated time: July 1, 2025 *** # April 04, 2025 Product Updates Source: https://novita.ai/docs/changelog/31-03-25--04-04-25 ## **New Features** * **Added Support for Hugging Face Login** Users can now sign in using their Hugging Face account. Social login now supports Hugging Face in addition to Google and GitHub. Updated time: April 02, 2025 * **Added Qwen2.5 VL 72B Instruct to the Model Library** Qwen2.5 VL 72B Instruct is the latest high-performance multimodal large language model launched by Novita. Featuring enhanced visual comprehension capabilities, it also demonstrates strong performance in complex reasoning and decision-making tasks. Updated time: March 31, 2025 * **Added Gemma-3-27b-it to the Model Library** Gemma-3-27b-it is the latest high-performance multimodal large language model launched by Novita. Built on Gemini 2.0 architecture, it possesses cross-modal reasoning capabilities across text, images, and short videos. Updated time: March 31, 2025 # July 31, 2025 Notice Source: https://novita.ai/docs/changelog/31-07-25 ### **Model Decommissioning Notice** To optimize resources and enhance the overall user experience, Novita will decommission the **Stable Diffusion 3 API** on **August 15, 2025 (UTC 0:00).** The API will be inaccessible after this date. Please update your applications before then to prevent service interruptions. # LLM FAQs Source: https://novita.ai/docs/guides/LLM-FAQ ## Voucher Usage Guide ### What types of voucher does Novita provide? Novita currently offers the following types of vouchers to help users experience the platform's services: * **New User Voucher**: Available to first-time registered users for trying out services of various types of models. * **Referral Voucher**: Earned by inviting others to register and complete tasks on the platform. ## Rate Limits for Model Usage ### What are the RPM limits for different users? RPM (Requests Per Minute) limits vary based on a user's verification level and account status. For detailed limits, please refer to the official documentation:
| Status | Description |
|---|---|
| VALIDATING | The input file is being validated before the batch can begin |
| PROGRESS | Batch is in progress |
| COMPLETED | Batch processing completed successfully |
| FAILED | Batch processing failed |
| EXPIRED | Batch exceeded deadline |
| CANCELLING | Batch is being cancelled |
| CANCELLED | Batch was cancelled |
2. The maximum input file size per batch is 100MB. ### Error Handling Errors encountered during batch processing are recorded in a separate error file, accessible via the error\_file\_id field. Common error codes include:
| Error Code | Description | Solution |
|---|---|---|
| 400 | Invalid request format | Check JSONL syntax and required fields |
| 401 | Authentication failed | Verify API key |
| 404 | Batch not found | Check batch ID |
| 429 | Rate limit exceeded | Reduce request frequency |
| 500 | Server error | Contact us |
> **Status**: Supported (OpenAI-Compatible) ## 1. Overview **Interleaved Thinking** is an advanced reasoning framework that enables models to perform explicit reasoning steps between tool calls. Models with Interleaved Thinking can: * Reflect on the current environment and tool outputs * Decide the next action based on updated reasoning * Maintain a continuous reasoning chain across multiple tool invocations * Provide transparent, inspectable multi-step thinking through `reasoning_details` or `reasoning_content` This capability transforms traditional function-calling into **agent-level tool use**, making complex workflows more accurate, reliable, and context-aware. Novita fully supports Interleaved Thinking for all models that natively expose reasoning streams (e.g., MiniMax-M2 and other OpenAI-compatible reasoning models). ## 2. Key Concepts ### 2.1 Interleaving Instead of executing a single reasoning phase followed by a tool call, the model performs: ```BASH theme={"system"} Reason → Tool Call → Observe → Reason → Tool Call → ... ``` This allows the model to adjust its strategy dynamically based on previous tool outputs. ### 2.2 Reasoning Details (`reasoning_details`) For some models, the content of the model's thinking will be returned in the form of a separate structure: ```JSON theme={"system"} "reasoning_details": [ { "type": "reasoning.text", "format": "openai-responses-v1", "text": "Model’s step-by-step reasoning..." } ] ``` For such models, Novita supports returning this field in both streaming and non-streaming modes. ### 2.3 Conversation Memory Requirement To maintain reasoning continuity: You **must append the model’s full response** including `reasoning_details`, `tool_calls`, and `content` to subsequent `messages`. Failing to preserve the chain may result in: * Incorrect tool use * Lost reasoning context * Repeated or circular tool calls * Reduced reliability This requirement mirrors OpenAI’s reasoning APIs. ## 3. API Behavior ### 3.1 Request Format No changes are required on the user side. Interleaved Thinking works with the standard OpenAI-compatible Chat Completions API. ### 3.2 Response Format The model may return the following fields: * `reasoning_content`: original thinking content * `reasoning_details`: structured reasoning segments, this field is optioanl * `tool_calls`: tool invocation plan * `content`: natural language output These extend the standard OpenAI format. ## 4. Example Request (MiniMax-M2) ```JSON theme={"system"} { "model": "minimax/minimax-m2", "messages": [ { "role": "user", "content": "How's the weather in San Francisco?" }, { "role": "assistant", "name": "MiniMax AI", "content": "", "tool_calls": [ { "id": "call_function_asqvfevfc8af_1", "type": "function", "function": { "name": "get_weather", "arguments": "{\"location\": \"San Francisco, US\"}" } } ], "reasoning_details": [ { "type": "reasoning.text", "id": "reasoning-text-1", "format": "openai-responses-v1", "index": 0, "text": "The user is asking about the weather in San Francisco..." } ] }, { "role": "tool", "tool_call_id": "call_function_asqvfevfc8af_1", "content": "24℃, sunny" } ], "stream": true, "reasoning_split": true, "max_tokens": 1024, "temperature": 0.7, "tools": [ { "type": "function", "function": { "name": "get_weather", "description": "Get weather for a specific location", "parameters": { "type": "object", "properties": { "location": { "type": "string" } }, "required": ["location"] } } } ] } ``` ## 5. Example Response (Non-Streaming) ```JSON theme={"system"} { "id": "07a4dedfdb1498b045498dfd42497639", "object": "chat.completion", "created": 1764303147, "model": "MiniMax", "choices": [ { "index": 0, "message": { "role": "assistant", "content": "", "name": "MiniMax", "tool_calls": [ { "index": 0, "id": "call_function_9w7wq1j9zmpl_1", "type": "function", "function": { "name": "get_weather", "arguments": "{\"location\": \"ShangHai\"}" } } ], "reasoning_content": "The user asked for Shanghai weather...", "reasoning_details": [ { "type": "reasoning.text", "text": "The user is asking about the weather in Shanghai...", "id": "reasoning-text-1", "format": "openai-responses-v1", "index": 0 } ] }, "finish_reason": "tool_calls" } ] } ``` ## 6. Streaming Response Example ```JSON theme={"system"} { "id": "664c5ad870c1888fbcbd267d9829e354", "object": "chat.completion.chunk", "created": 1764303504, "model": "minimax-m2", "choices": [ { "index": 0, "delta": { "role": "assistant", "reasoning_content": "...\n\nThe user has specifically asked...", "reasoning_details": [ { "type": "reasoning.text", "text": ".\n\nThe user has specifically asked...", "id": "reasoning-text-1", "format": "openai-responses-v1", "index": 0 } ] }, "finish_reason": null } ] } ``` ## 7. Developer Notes ### 7.1 Models That Support Interleaved Thinking All models exposing `reasoning_details` through OpenAI-compatible APIs, including: * MiniMax-M2 * (Upcoming) Novita Reasoning Series * Other reasoning-enabled partner models ### 7.2 Pricing Billing is based on reasoning tokens, following the model’s pricing rules. `reasoning_details` will increase token usage. ### 7.3 Error Handling You may encounter: * Missing tool parameters * Recursive or repeated tool calls * Incorrect assumptions in the reasoning phase Ensure your application validates tool arguments and handles model errors gracefully. ## 8. Best Practices **✓ Always include full model messages in the next request** Include: * `content` * `tool_calls` * `reasoning_details` **✓ Enable streaming for long-chain reasoning** Streaming allows the client to: * Monitor the reasoning process * Detect incorrect tool plans early * Provide faster user feedback **✓ Combine with server-side state machines for stability** For production systems, we recommend pairing Interleaved Thinking with deterministic guardrails, e.g.: * Parameter validators * Execution sandbox * Maximum recursion safeguards ## 9. Summary Interleaved Thinking significantly enhances multi-step reasoning and tool-use reliability: * Transparent and inspectable step-wise reasoning * Adaptive planning between tool invocations * Stronger context retention across long workflows * Fully compatible with OpenAI-style Chat Completions API Novita will continue expanding support for advanced reasoning models, bringing agent-level intelligence to the API layer. # LLM Monitoring Source: https://novita.ai/docs/guides/llm-monitor Description of your new file. Novita AI provides comprehensive monitoring metrics for the usage of large language model (LLM) APIs. These metrics help you gain deep insights into the availability and performance of your LLM API requests. You can view the monitoring metrics on the [LLM Monitoring Page](https://novita.ai/models-console/llm-metrics). ## Metric Descriptions
| Tier | How to reach |
|---|---|
| T1 | Monthly top-ups did not exceed \$50 in any of the last 3 calendar months. |
| T2 | Monthly top-ups were at least \$50 but did not exceed \$500 in any of the last 3 calendar months. |
| T3 | Monthly top-ups were at least \$500 but did not exceed \$3,000 in any of the last 3 calendar months. |
| T4 | Monthly top-ups were at least \$3,000 but did not exceed \$10,000 in any of the last 3 calendar months. |
| T5 | Monthly top-ups were at least \$10,000 in at least one of the last 3 calendar months. |
[**Qwen-2.5-72b-instruct**](https://novita.ai/models/llm/qwen-qwen-2.5-72b-instruct), [**Llama-3.3-70b-instruct**](https://novita.ai/models/llm/meta-llama-llama-3.3-70b-instruct) *(Medium)* | | **Function calling & tool use** | [**Qwen3-235b-a22b-fp8**](https://novita.ai/models/llm/qwen-qwen3-235b-a22b-fp8) *(Large)*
**Qwen 3 Family Models** *(Large/Medium/Small)* | | **Long context & summarization** | [**Llama-4-maverick-17b-128e-instruct-fp8**](https://novita.ai/models/llm/meta-llama-llama-4-maverick-17b-128e-instruct-fp8) *(Large)*
[**Llama-4-scout-17b-16e-instruct**](https://novita.ai/models/llm/meta-llama-llama-4-scout-17b-16e-instruct) *(Medium)* | | **Vision & document understanding** | [**Llama-4-maverick-17b-128e-instruct-fp8**](https://novita.ai/models/llm/meta-llama-llama-4-maverick-17b-128e-instruct-fp8) *(Large)*
[**Qwen2.5-vl-72b-instruct**](https://novita.ai/models/llm/qwen-qwen2.5-vl-72b-instruct), [**Llama-4-scout-17b-16e-instruct**](https://novita.ai/models/llm/meta-llama-llama-4-scout-17b-16e-instruct) *(Medium)* | | **Low-latency NLU & extraction** | [**Llama-3.1-8b-instruct**](https://novita.ai/models/llm/meta-llama-llama-3.1-8b-instruct), [**Llama-3.2-3b-instruct**](https://novita.ai/models/llm/meta-llama-llama-3.2-3b-instruct), [**Llama-3.2-1b-instruct**](https://novita.ai/models/llm/meta-llama-llama-3.2-1b-instruct) *(Small)* | *Last updated: August 1, 2025* # Structured Outputs Source: https://novita.ai/docs/guides/llm-structured-outputs `Structured Outputs` enables the model to generate responses that adhere to your supplied [JSON Schema](https://json-schema.org/specification). ## Supported Models The following models support `Structured Outputs`:
| V2 | V3 | Description |
|---|---|---|
| **extra**object | **extra**object | |
| enable\_nsfw\_detection boolean |
enable\_nsfw\_detection boolean |
|
| nsfw\_detection\_level Enum: `0, 1, 2` |
nsfw\_detection\_level Enum: `0, 1, 2` |
|
| enable\_progress\_info | Deprecated | |
| response\_image\_type Enum: `png`, `jpeg` |
response\_image\_type Enum: `png, webp, jpeg` |
V3 adds support for `webp`image format |
| **request**object | New Field All image generation parameters must be passed via the `request`in V3 |
|
| **prompt**string \ |
promptstring | Moved Inside |
| lorasobject\[] | Moved Inside **Migrate LoRA usage: From **`prompt`** to **`request.loras`** parameter** |
|
| model\_namestring | New Field Name of lora, retrieve the corresponding sd\_name\_in\_api value by invoking the [Get Model API](https://novita.ai/docs/api-reference/model-apis-get-model) endpoint with filter.types=lora as the query parameter. |
|
| strengthnumber(float32) | New Field The strength value of lora. The larger the value, the more biased the effect is towards lora, Range \[0, 1] |
|
| **negative\_prompt**string | negative\_promptstring | Moved Inside |
| **sampler\_name**string | sampler\_namestring | Moved Inside |
| **batch\_size**integer | image\_numinteger | Changed `num_images` **→** `request.image_num` |
| **n\_iter** | Deprecated | |
| **steps**string | stepsstring | Moved Inside |
| **cfg\_scale**integer | guidance\_scale number(float32) |
Changed `cfg_scale`**→** `request.guidance_scale` |
| **seed**integer | seedinteger | Moved Inside |
| **height**integer | heightinteger | Moved Inside Range Change: \[128, 2048]. |
| **width**integer | widthinteger | Moved Inside Range Change: \[128, 2048]. |
| **model\_name**string | model\_namestring | Moved Inside This parameter specifies the name of the model checkpoint. Retrieve the corresponding sd\_name value by invoking the [Query Model](https://novita.ai/docs/api-reference/model-apis-get-model) API with filter.types=checkpoint as the query parameter. |
| **restore\_faces**bool | restore\_facesbool | Moved Inside |
| **restore\_faces\_model** | Deprecated | |
| **sd\_vae**string | sd\_vaestring | Moved Inside |
| **clip\_skip**integer | clip\_skipinteger | Moved Inside |
| **enable\_hr**boolean | hires\_fixobject | Changed `enable_hr`**→** `request.hires_fix` |
| **hr\_upscaler** Enum: `Latent`, `ESRGAN_4x`, `R-ESRGAN 4x+`, `R-ESRGAN 4x+ Anime6B` |
upscaler Enum: `RealESRGAN_x4plus_anime_6B`, `RealESRNet_x4plus,Latent` |
Changed `hr_upscaler`**→**`request.hires_fix.upscaler` |
| **hr\_scale**number | Deprecated | |
| **hr\_resize\_x**integer | target\_widthinteger | Changed `hr_resize_x`**→** `request.hires_fix.target_width` |
| **hr\_resize\_y**integer | target\_heightinteger | Changed `hr_resize_y`**→** `request.hires_fix.target_height` |
| **img\_expire\_ttl**integer | Deprecated Default 3600s |
|
| **sd\_refiner**object | refinerobject | Changed `sd_refiner`**→** `request.refiner` |
| checkpointstring | Deprecated | |
| switch\_at number(float32) |
switch\_at number(float32) |
Changed `sd_refiner.switch_at`**→** `request.refiner.switch_at` |
| **controlnet\_units**object\[] | Deprecated `img2img` Only |
| V2 | V3 | Description |
|---|---|---|
| **code** | Deprecated | |
| **msg** | Deprecated | |
| **data** | Deprecated | |
| task\_id | **task\_id** | Changed `data.task_id`**→** `task_id` |
| warn | Deprecated |
| V2 | V3 | Description |
|---|---|---|
| **extra**object | **extra**object | |
| enable\_nsfw\_detection boolean |
enable\_nsfw\_detection boolean |
|
| nsfw\_detection\_level Enum: `0, 1, 2` |
nsfw\_detection\_level Enum: `0, 1, 2` |
|
| enable\_progress\_info | Deprecated | |
| response\_image\_type Enum: `png`, `jpeg` |
response\_image\_type Enum: `png, webp, jpeg` |
V3 adds support for `webp`image format |
| **request**object | New field All image generation parameters must be passed via the `request`in V3 |
|
| **prompt**string \ |
promptstring | Moved Inside |
| lorasobject\[] | Moved Inside **Migrate LoRA usage: From **`prompt`** to **`request.loras`** parameter** |
|
| model\_namestring | New Field Name of lora, retrieve the corresponding sd\_name\_in\_api value by invoking the [Get Model API](https://novita.ai/docs/api-reference/model-apis-get-model) endpoint with filter.types=lora as the query parameter. |
|
| strength number(float32) |
New Field The strength value of lora. The larger the value, the more biased the effect is towards lora, Range \[0, 1] |
|
| **negative\_prompt**string | negative\_prompt string |
Moved Inside |
| **sampler\_name**string | sampler\_namestring | Moved Inside |
| **batch\_size**integer | image\_numinteger | Changed `batch_size`**→** `request.image_num` |
| **n\_iter**integer | Deprecated | |
| **steps**string | stepsstring | Moved Inside |
| **cfg\_scale**integer | guidance\_scale number(float32) |
Changed `cfg_scale`**→** `request.guidance_scale` |
| **seed**integer | seedinteger | Moved Inside |
| **height**integer | heightinteger | Moved Inside Range Change: \[128, 2048]. |
| **width**integer | widthinteger | Moved Inside Range Change: \[128, 2048]. |
| **model\_name**string | model\_namestring | Moved Inside This parameter specifies the name of the model checkpoint. Retrieve the corresponding sd\_name value by invoking the [Query Model](https://novita.ai/docs/api-reference/model-apis-get-model) API with filter.types=checkpoint as the query parameter. |
| **init\_images**string\[] | image\_base64string | Changed `init_images`**→** `request.image_base64` |
| **denoising\_strength** number(float) |
strength number(float) |
Changed `denoising_strength`**→** `request.strength` |
| **restore\_faces**bool | Deprecated | |
| **sd\_vae**string | sd\_vaestring | Moved Inside |
| **clip\_skip**integer | clip\_skipinteger | Moved Inside |
| **mask**string | Deprecated Recommendation: Use V3 Inpainting API |
|
| **mask\_blur**integer | Deprecated Recommendation: Use V3 Inpainting API |
|
| **resize\_mode**integer | Deprecated | |
| **image\_cfg\_scale**integer | Deprecated | |
| **inpainting\_fill**integer | Deprecated Recommendation: Use V3 Inpainting API |
|
| **inpaint\_full\_res**integer | Deprecated Recommendation: Use V3 Inpainting API |
|
| **inpaint\_full\_res\_padding** integer |
Deprecated Recommendation: Use V3 Inpainting API |
|
| **inpainting\_mask\_invert** integer |
Deprecated Recommendation: Use V3 Inpainting API |
|
| **initial\_noise\_multiplier** number(float32) |
Deprecated | |
| **img\_expire\_ttl**integer | Deprecated Default 3600s |
|
| **sd\_refiner**object | refinerobject | Changed `sd_refiner`**→** `request.refiner` |
| checkpoint | Deprecated | |
| switch\_at number(float32) |
switch\_at number(float32) |
Moved Inside |
| controlnetobject | New Field | |
| **controlnet\_units**object\[] | unitsobject\[] | Changed `controlnet_units`**→** `request.controlnet.units` |
| modelstring | model\_name string |
Changed `controlnet_units.model`**→** `request.controlnet.units.model_name` |
| weightnumber | strength number(float32) |
Changed `controlnet_units.weight`**→** `request.controlnet.units.strength` |
| input\_imagestring | image\_base64 string |
Changed `controlnet_units.input_image`**→** `request.controlnet.units.image_base64` |
| modulestring,Enum | preprocessor string,Enum |
Changed `controlnet_units.module`**→** `request.controlnet.units.preprocessor` |
| control\_mode | Deprecated | |
| mask | Deprecated Recommendation: Use V3 Inpainting API |
|
| resize\_mode | Deprecated | |
| processor\_res | Deprecated | |
| threshold\_a | Deprecated | |
| threshold\_b | Deprecated | |
| guidance\_start number(float32) |
guidance\_start number(float32) |
Moved Inside |
| guidance\_end number(float32) |
guidance\_end number(float32) |
Moved Inside |
| pixel\_perfect | Deprecated |
| V2 | V3 | Description |
|---|---|---|
| code | Deprecated | |
| msg | Deprecated | |
| data | Deprecated | |
| task\_id | **task\_id** | Changed `data.task_id`**→** `task_id` |
| warn | Deprecated |
Directory listing for /
////
-
//
- .bash_logout //
- .bashrc //
- .profile //
// // await sandbox.kill() ``` ```python Python icon="python" theme={"system"} from novita_sandbox.code_interpreter import Sandbox sandbox = Sandbox.create() # Start a simple HTTP server inside the sandbox. command_handle = sandbox.commands.run("python -m http.server 3000", background=True) host = sandbox.get_host(3000) url = f"https://{host}" print('Server started at:', url) # Fetch data from the server inside the sandbox. response = sandbox.commands.run(f"curl {url}") data = response.stdout print("Response from server inside sandbox:", data) # Kill the server process inside the sandbox. command_handle.kill() # Example output: # Server started at: https://3000-i0iq56w6786bz91034h8l-82b888ba.sandbox.novita.ai # Response from server inside sandbox: # # # #
Directory listing for /
##
-
#
- .bash_logout #
- .bashrc #
- .profile #
# # sandbox.kill() ```
| Billing Item | Description |
|---|---|
| CPU | Billed based on the number of vCPU cores used and usage duration (accurate to the second). Current pricing can be found on the [here](https://novita.ai/pricing?sandbox=1). No billing occurs after the Sandbox is stopped. |
| RAM (memory) | Billed based on allocated memory capacity and usage duration (accurate to the second). Current pricing can be found on the [here](https://novita.ai/pricing?sandbox=1). No billing occurs after the Sandbox is stopped. |
| Storage | Current pricing can be found on the [here](https://novita.ai/pricing?sandbox=1). |
| Templates | Currently free. |
| vCPUs (cores) | Unit Price |
|---|---|
| 1 | \$0.0000098/s |
| 2 | \$0.0000196/s |
| 3 | \$0.0000294/s |
| 4 | \$0.0000392/s |
| 5 | \$0.000049/s |
| 6 | \$0.0000588/s |
| 7 | \$0.0000686/s |
| 8 | \$0.0000784/s |
| Memory (MiB) | Unit Price |
|---|---|
| Valid values: multiples of 512 MiB, from 512 MiB to 8192 MiB | \$0.0000032/GiB/s |
| 512 MiB | \$0.0000016/s |
| 1 GiB | \$0.0000032/s |
| 2 GiB | \$0.0000064/s |
| Memory (MiB) | Unit Price |
|---|---|
| Each account includes 60 GB of free storage. Usage beyond this limit is billed on a pay-as-you-go basis. | \$0.00009/GB/h |
| Configuration Item | Description |
|---|---|
| Min Worker Count | The minimum number of worker instances to keep for the endpoint. Setting a higher minimum helps reduce cold start time. If set to 0, there will be no idle workers when there are no requests, which may increase response time for incoming requests. For latency-sensitive scenarios, use 0 with caution. |
| Max Worker Count | The maximum number of worker instances that the endpoint can scale up to. When request volume increases, the platform automatically increases workers up to this maximum. This limit helps control costs. |
| Idle Timeout (seconds) | When a worker is about to be released due to autoscaling down, the platform will keep the worker alive for the specified idle timeout period to be able to react quickly to new requests. Note that you will be charged for the worker during this period. |
| Max Concurrent Requests | The maximum number of concurrent requests handled by a single worker. If this is exceeded, requests will be routed to other workers. If all workers are fully occupied, excess requests will be queued until execution is possible. |
| GPUs / Worker | Number of GPU cards allocated to each worker. |
| CUDA Version | Specify the CUDA version supported for the worker. |