Model APIs Updates

New Features

  • New Model Available: DeepSeek-R1-0528

    Novita now supports the latest high-performance DeepSeek model — DeepSeek-R1-0528.

    Released by the DeepSeek team as an open-source model, DeepSeek-R1-0528 features impressive reasoning capabilities, particularly achieving performance comparable to OpenAI’s o1 model in mathematics, coding, and reasoning tasks.

    Updated time: May 28, 2025

Discontinued Features

  • Certain LLMs Officially Decommissioned

    To continuously improve model performance and enhance user experience, several large language models (LLMs) have now been officially decommissioned. Below is the list of retired models along with their recommended replacements:

    Deprecated ModelsDeprecation DateReplacement Models
    jondurbin/airoboros-l2-70b2025-05-28sao10k/l3-70b-euryale-v2.1
    qwen/qwen3-14b-fp82025-05-28qwen/qwen3-32b-fp8
    qwen/qwen3-0.6b-fp82025-05-28qwen/qwen3-32b-fp8
    qwen/qwen3-1.7b-fp82025-05-28qwen/qwen3-32b-fp8
    meta-llama/llama-3.2-11b-vision-instruct2025-05-28qwen/qwen2.5-vl-72b-instruct
    google/gemma-2-9b-it2025-05-28google/gemma-3-27b-it
    meta-llama/llama-3.1-70b-instruct2025-05-28meta-llama/llama-3.3-70b-instruct
    qwen/qwq-32b2025-05-28meta-llama/llama-3.3-70b-instruct

    If you encounter any issues during this transition, please don’t hesitate to contact our technical support team.

    Updated time: May 28, 2025


GPUs Updates

New Features

  • New Bare Metal Server Reservation Page Launched

    Novita has launched a bare metal server reservation interface on the official website. Users can now browse available GPU models that support bare metal services, along with detailed configuration information.

    Bare metal refers to a computing service that provides direct access to physical servers without a virtualization layer, offering superior performance, greater control, and lower latency. It is ideal for scenarios with high demands on resource isolation and computing power, such as deep learning, AI model training, and large-scale inference.

    If you require bare metal resources, you can select your preferred GPU model and complete the reservation online to gain exclusive access to dedicated physical servers.

    Updated time: May 30, 2025