Cheapest LLM API for Each Current Open Model

At list prices, DeepInfra is cheapest for a 10,000-input/1,000-output request on GLM-5.3 ($0.01300) and DeepSeek V4.1 Flash 1017 ($0.00260), while seven providers tie on GLM-5.3-Flash at $0.00200. Novita AI is not the cheapest on any of the three: it charges the standard rates most hosts share.

Updated · 22 sources · Methodology

NeedPickWhy
Cheapest GLM-5.3DeepInfra$0.01300 per example at list price, vs $0.01840 at Novita AI84
Cheapest DeepSeek V4.1 Flash 1017DeepInfra$0.00260 per example at list price, vs $0.00420 at Novita AI95
Cheapest GLM-5.3-FlashSeven-way tie$0.00200 per example at every provider listed1516101461217
Cheapest GPT-OSS 120B (baseline)CoreWeave$0.00047 per example, vs $0.00075 at Novita AI167

GLM-5.3: request cost by provider

ProviderInput / 1M tokensOutput / 1M tokensEstimated request cost
DeepInfra8$0.90$4.00$0.01300
Baseten15$1.40$4.40$0.01840
Fireworks14$1.40$4.40$0.01840
Novita AI4$1.40$4.40$0.01840
Together AI12$1.40$4.40$0.01840
Z.ai17$1.40$4.40$0.01840
Baseten (Fast)15$2.10$6.60$0.02760
Cost example: 10,000 input + 1,000 billed output tokens
DeepInfra$0.01300
Baseten$0.01840
Fireworks$0.01840
Novita AI$0.01840
Together AI$0.01840
Z.ai$0.01840
Baseten (Fast)$0.02760

DeepInfra is shown at its list price; its current promotion is in its card below. CoreWeave does not list GLM-5.3 on its public price list.816

DeepSeek V4.1 Flash 1017: request cost by provider

ProviderInput / 1M tokensOutput / 1M tokensEstimated request cost
DeepInfra9$0.20$0.60$0.00260
Baseten15$0.30$1.20$0.00420
DeepSeek18$0.30$1.20$0.00420
Fireworks14$0.30$1.20$0.00420
Novita AI5$0.30$1.20$0.00420
Together AI12$0.30$1.20$0.00420
Baseten (Fast)15$0.60$2.40$0.00840
Cost example: 10,000 input + 1,000 billed output tokens
DeepInfra$0.00260
Baseten$0.00420
DeepSeek$0.00420
Fireworks$0.00420
Novita AI$0.00420
Together AI$0.00420
Baseten (Fast)$0.00840

DeepInfra is shown at its list price. DeepSeek’s row is its peak rate; off-peak it bills $0.15 input and $0.60 output, or $0.00210 per example. Baseten lists its Fast tier as DeepSeek V4.1 Flash Fast, without the 1017 suffix, and CoreWeave lists DeepSeek V4.1-Flash without the 1017 version, so CoreWeave is not included.9181516

GLM-5.3-Flash: request cost by provider

ProviderInput / 1M tokensOutput / 1M tokensEstimated request cost
Baseten15$0.15$0.50$0.00200
CoreWeave16$0.15$0.50$0.00200
DeepInfra10$0.15$0.50$0.00200
Fireworks14$0.15$0.50$0.00200
Novita AI6$0.15$0.50$0.00200
Together AI12$0.15$0.50$0.00200
Z.ai17$0.15$0.50$0.00200

All seven providers list identical input and output rates, so this request mix cannot separate them; speed and first-chunk wait still differ, as the cards show.

GPT-OSS 120B baseline

Kept from earlier editions of this page as a reference. Groq and Cerebras serve none of the three current models above, so GPT-OSS 120B is the only model here where their prices can be set side by side.2022

ProviderInput / 1M tokensOutput / 1M tokensEstimated request cost
CoreWeave16$0.03$0.17$0.00047
DeepInfra11$0.037$0.17$0.00054
Novita AI7$0.05$0.25$0.00075
Groq19$0.15$0.60$0.00210
Together AI13$0.15$0.60$0.00210
Cerebras21$0.35$0.75$0.00425

Options, strengths and limitations

Each card shows the provider’s list-price cost of the 10,000-input/1,000-output example on each current model.

  1. DeepInfra

    Cheapest list price on GLM-5.3 and DeepSeek V4.1 Flash 1017, and tied on GLM-5.3-Flash.891011123

    GLM-5.3
    $0.01300
    DeepSeek V4.1 Flash 1017
    $0.00260
    GLM-5.3-Flash
    $0.00200

    Strengths

    Against the standard hosts, its list rates cut the GLM-5.3 example by 29% ($0.01300 vs $0.01840) and the DeepSeek V4.1 Flash 1017 example by 38% ($0.00260 vs $0.00420); on GPT-OSS 120B it is second-cheapest at $0.00054.891011123

    Limitations

    It has the lowest output speed of the measured hosts on all three current models (97.1, 65.3 and 57.2 tokens/s), although its first chunk arrives in under 1 s on each.891011123

    Current promotion: GLM-5.3 at $0.5625 input and $2.50 output ($0.00813 per example) and DeepSeek V4.1 Flash 1017 at $0.098 and $0.294 ($0.00127). Promotions can end, so the tables use list prices.89

    891011123

  2. Baseten

    Standard list rates on all three current models, plus a Fast tier that costs more.151

    GLM-5.3
    $0.01840
    DeepSeek V4.1 Flash 1017
    $0.00420
    GLM-5.3-Flash
    $0.00200

    Strengths

    For the same $0.01840 GLM-5.3 example as Novita AI, Together AI, Fireworks and Z.ai, it streams 150.4 tokens/s, against 105.5 at Novita AI.151

    Limitations

    Its Fast tier costs 1.5× the standard GLM-5.3 example ($0.02760) and 2× on DeepSeek V4.1 Flash 1017 ($0.00840), the highest in both tables.151

    151

  3. Fireworks

    Standard list rates on all three current models, with no price advantage on any of them.1421

    GLM-5.3
    $0.01840
    DeepSeek V4.1 Flash 1017
    $0.00420
    GLM-5.3-Flash
    $0.00200

    Strengths

    On DeepSeek V4.1 Flash 1017 it pairs the standard $0.00420 example with 210.4 tokens/s, against 158.0 at Novita AI.1421

    Limitations

    The rates shown are its Standard global tier, and Artificial Analysis has no Fireworks row for GLM-5.3, so its speed there is unmeasured.1421

    1421

  4. Novita AI

    Not the cheapest on any current model: it charges the standard list rates and only ties on GLM-5.3-Flash.456789123

    GLM-5.3
    $0.01840
    DeepSeek V4.1 Flash 1017
    $0.00420
    GLM-5.3-Flash
    $0.00200

    Strengths

    It streams faster than DeepInfra, the cheapest host, on all three: 105.5 vs 97.1 tokens/s on GLM-5.3, 158.0 vs 65.3 on DeepSeek V4.1 Flash 1017 and 94.3 vs 57.2 on GLM-5.3-Flash.456789123

    Limitations

    DeepInfra’s list price is 29% lower on GLM-5.3 and 38% lower on DeepSeek V4.1 Flash 1017, its 4.00 s first chunk on GLM-5.3-Flash is the slowest of the measured hosts, and on GPT-OSS 120B it ranks third at $0.00075.456789123

    456789123

  5. Together AI

    Standard list rates on all three current models, with the shortest GLM-5.3 first-chunk wait in the table.12131612

    GLM-5.3
    $0.01840
    DeepSeek V4.1 Flash 1017
    $0.00420
    GLM-5.3-Flash
    $0.00200

    Strengths

    For the same $0.00420 DeepSeek V4.1 Flash 1017 example as Novita AI, it streams 365.0 tokens/s with a 0.32 s first chunk, versus 158.0 and 1.43 s; on GLM-5.3 its first chunk takes 0.67 s.12131612

    Limitations

    On GPT-OSS 120B its $0.00210 example is 4.5× CoreWeave’s $0.00047.12131612

    12131612

  6. Z.ai

    The model creator’s own API charges the same list rates as the standard hosts on GLM-5.3 and GLM-5.3-Flash.1713

    GLM-5.3
    $0.01840
    DeepSeek V4.1 Flash 1017
    Not listed
    GLM-5.3-Flash
    $0.00200

    Strengths

    Buying first-party costs nothing extra: its GLM-5.3 example matches the $0.01840 standard, and GLM-5.3-Flash matches the $0.00200 tie.1713

    Limitations

    It is among the slower GLM hosts measured, ahead only of DeepInfra: 98.2 tokens/s on GLM-5.3 and 73.5 on GLM-5.3-Flash.1713

    1713

  7. CoreWeave

    Cheapest on the GPT-OSS 120B baseline at $0.00047 and tied on GLM-5.3-Flash, but absent from the other two current models.163

    GLM-5.3
    Not listed
    DeepSeek V4.1 Flash 1017
    Not listed
    GLM-5.3-Flash
    $0.00200

    Strengths

    At the tied $0.00200 price it is the fastest GLM-5.3-Flash host: 292.4 tokens/s, against 190.1 at Together AI.163

    Limitations

    Its public price list has no GLM-5.3 and only an earlier DeepSeek V4.1-Flash, so it cannot be priced on two of the three current models.163

    163

  8. DeepSeek

    The first-party API for DeepSeek V4.1 Flash 1017 matches the standard $0.00420 example at peak rates.189

    GLM-5.3
    Not listed
    DeepSeek V4.1 Flash 1017
    $0.00420
    GLM-5.3-Flash
    Not listed

    Strengths

    Off-peak rates of $0.15 input and $0.60 output cut the example to $0.00210, below DeepInfra’s $0.00260 list price, for workloads that can run in those hours.189

    Limitations

    It serves only its own model here, and at peak it is no cheaper than Novita AI, Together AI, Fireworks or Baseten.189

    189

Methodology and sources

Prices are each provider’s official standard serverless rates per 1M tokens. Where a provider is running a promotion (DeepInfra on GLM-5.3 and DeepSeek V4.1 Flash 1017), tables and rankings use the list price and the provider’s card gives the promotional rate. Each model is ranked on its own: speed and first-chunk figures in the cards are Artificial Analysis medians at 10,000 input tokens, and AA’s reasoning-effort setting differs by model, so figures are not comparable across models.

Each request cost is the list input rate × 10,000 plus the output rate × 1,000, divided by 1M. Billed output includes reasoning tokens; cache discounts, taxes, retries and negotiated contracts are left out. Novita AI publishes this page from public price lists and benchmarks and ran no private workload.

Cards follow the GLM-5.3 list-price example, with tied providers in alphabetical order; CoreWeave and DeepSeek, which list no GLM-5.3 price, come last. GPT-OSS 120B is a baseline only and does not affect the order.

  1. Artificial Analysis · GLM-5.3 Providers
  2. Artificial Analysis · DeepSeek V4.1 Flash 1017 Providers
  3. Artificial Analysis · GLM-5.3-Flash Providers
  4. Novita AI · GLM-5.3
  5. Novita AI · DeepSeek V4.1 Flash 1017
  6. Novita AI · GLM-5.3-Flash
  7. Novita AI · GPT-OSS 120B
  8. DeepInfra · GLM-5.3
  9. DeepInfra · DeepSeek V4.1 Flash 1017
  10. DeepInfra · GLM-5.3-Flash
  11. DeepInfra · GPT-OSS 120B
  12. Together AI · model API prices
  13. Together AI · GPT-OSS 120B
  14. Fireworks · serverless pricing
  15. Baseten · pricing
  16. CoreWeave · Forge token prices
  17. Z.ai · pricing
  18. DeepSeek · API pricing
  19. Groq · GPT-OSS 120B
  20. Groq · supported models
  21. Cerebras · pricing
  22. Cerebras · supported models

Build with Novita Model APIs

Create an API key and connect your application.

Start BuildingExplore Models