Sign in

Models and pricing

Every model, priced per token.

Input and output rates for each model, with its context window. Batch work is billed per finished line, after its result is stored.

USD per 1 million tokens. Context is the window we serve.

ModelContextInputCached inputOutput
gpt-oss-20bmxfp4openai/gpt-oss-20bLicense: Apache-2.0131K$0.03–$0.15Start a batch →
Coming soon
DeepSeek V4.1 FlashFP8DeepSeek · deepseek-v4.1-flashReasoningVisionToolsJSON1M$0.07$0.003$0.49Coming soon
DeepSeek V4.1 Flash CyberFP8DeepSeek · deepseek-v4.1-flash-cyberReasoningVisionToolsJSON1M$0.30$0.006$1.00Coming soon
Gemma 4 26B A4B ITBF16Google · gemma-4-26b-a4b-itReasoningVisionToolsJSON256K$0.11$0.05$0.37Coming soon
Gemma 4 31B ITBF16Google · gemma-4-31b-itReasoningVisionToolsJSON256K$0.14$0.05$0.40Coming soon
GLM-5.3FP8Z.ai · glm-5.3ReasoningToolsJSON230K$0.70$0.13$2.20Coming soon
GLM-5.3 FlashFP8Z.ai · glm-5.3-flashReasoningToolsJSON1M$0.075$0.015$0.25Coming soon
Nemotron 3.5 Super VL 120B A12BNVFP4NVIDIA · nemotron-3.5-super-vl-120b-a12bReasoningVisionToolsJSON1M$0.15$0.02$0.50Coming soon
Qwen 3.8 27BFP8Alibaba · qwen-3.8-27bReasoningVisionToolsJSON256K$0.30$0.08$2.35Coming soon

Prices of models marked Coming soon are indicative and subject to change.

These prices are for batch work. Sign in to see the current prices.

How billing works

  • You pay first. Your credit is prepaid, through Stripe.
  • You are charged for each finished line only.
  • The unused part of a batch's hold is released when the batch ends.
  • Each top-up is from $25 to $500.
  • A batch has a window of 24 hours.

Questions

How is a batch charged?

Each finished line is charged once, after its result is stored, at the price of its model per 1 million tokens. A line that ends in an error is not charged.

When is the hold released?

When you create a batch, we hold its estimated cost from your credit. When the batch ends, we charge the finished lines and release the rest of the hold.

What happens when a line fails?

The line goes to the error file of the batch, and you are not charged for it. The other lines of the batch still finish.

Do you store my prompts?

We use your files only to run your batches. Our logs hold IDs, counts, timings and error codes, and never the text of a prompt or a completion.

Which OpenAI SDK calls work?

For files: create, list, retrieve, download the content and delete. For batches: create, list, retrieve and cancel. A request line uses the /v1/chat/completions endpoint.

What is the batch window?

24 hours. If a batch does not finish in that time, it expires. You get the results of the lines that finished, and you pay only for those lines.

Is there a charge when I run no batch?

No. You pay for finished lines only. Credit that you do not use stays on your account.

How much credit can I add?

You add credit through Stripe, from $25 to $500 each time.

Ready to run a batch?

Start a batch