Inference
Models and pricing
Token pricing
Prices are in USD per million tokens. Cached input is charged at the cached rate instead of the regular input rate. There is no separate cache-write fee.
| Model ID | Input | Cached input | Output |
|---|---|---|---|
Capabilities and limits
All models support streaming, reasoning, and function calling. Output is also bounded by the remaining context after input.
| Model ID | Context (input + output) | Maximum output |
|---|---|---|
moonshotai/Kimi-K3.1-Pro | 1,048,576 tokens | 1,048,575 tokens |
moonshotai/Kimi-K3 | 1,048,576 tokens | 1,048,575 tokens |
deepseek-ai/DeepSeek-V4.1-Flash-1017 | 1,048,576 tokens | 393,216 tokens |
deepseek-ai/DeepSeek-V4.1-Flash | 1,048,576 tokens | 393,216 tokens |
nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4 | 1,048,576 tokens | 1,048,575 tokens |
zai-org/GLM-5.3-Flash | 1,048,576 tokens | 128K tokens |
Qwen/Qwen3.8-27B-FP8 | 262,144 tokens | 262,143 tokens |
GLM-5.3 | 1,048,576 tokens | 128K tokens |
See the quickstart to make your first request.
Usage and billing
| Token count | Chat Completions usage | Responses usage |
|---|---|---|
| All input tokens, including cached input | prompt_tokens | input_tokens |
| Cached subset of input | prompt_tokens_details.cached_tokens | input_tokens_details.cached_tokens |
| All generated tokens, including reasoning | completion_tokens | output_tokens |
| Reasoning subset of output | completion_tokens_details.reasoning_tokens | output_tokens_details.reasoning_tokens |