Serverless Inference

Pricing

Pay only for the tokens you use, with no platform fee or minimum. Every chat model is billed per million input and output tokens, and repeated prompt prefixes bill at a discounted cache-read rate.

/
Features
deepseek4-flashDeepSeek V4 Flash$0.11$0.66$0.007ChatMulti-turn conversations via /v1/chat/completions.StreamingTokens arrive as they are generated with stream: true.Tool callingNative tool calls, including parallel calls.ReasoningThinks before answering; the trace is billed as output.+3Structured outputsJSON Schema responses; JSON mode on most models.LogprobsReturns token log-probabilities with logprobs: true.Prompt cachingRepeated prompt prefixes bill at the cache-read rate.
deepseek-v4-proDeepSeek V4 Pro$1.20$3.30$0.13ChatMulti-turn conversations via /v1/chat/completions.StreamingTokens arrive as they are generated with stream: true.Tool callingNative tool calls, including parallel calls.ReasoningThinks before answering; the trace is billed as output.+3Structured outputsJSON Schema responses; JSON mode on most models.LogprobsReturns token log-probabilities with logprobs: true.Prompt cachingRepeated prompt prefixes bill at the cache-read rate.
glm5.3GLM 5.3$1.17$3.96$0.234ChatMulti-turn conversations via /v1/chat/completions.StreamingTokens arrive as they are generated with stream: true.Tool callingNative tool calls, including parallel calls.ReasoningThinks before answering; the trace is billed as output.+3Structured outputsJSON Schema responses; JSON mode on most models.LogprobsReturns token log-probabilities with logprobs: true.Prompt cachingRepeated prompt prefixes bill at the cache-read rate.
glm5.3-flashGLM 5.3 Flash$0.15$0.50$0.03ChatMulti-turn conversations via /v1/chat/completions.StreamingTokens arrive as they are generated with stream: true.Image inputAccepts images alongside text in the prompt.Tool callingNative tool calls, including parallel calls.+4ReasoningThinks before answering; the trace is billed as output.Structured outputsJSON Schema responses; JSON mode on most models.LogprobsReturns token log-probabilities with logprobs: true.Prompt cachingRepeated prompt prefixes bill at the cache-read rate.
qwen3.8-27bQwen3.8 27B$0.20$2.50$0.05ChatMulti-turn conversations via /v1/chat/completions.StreamingTokens arrive as they are generated with stream: true.Image inputAccepts images alongside text in the prompt.Tool callingNative tool calls, including parallel calls.+3ReasoningThinks before answering; the trace is billed as output.Structured outputsJSON Schema responses; JSON mode on most models.Prompt cachingRepeated prompt prefixes bill at the cache-read rate.
reka-edge-2603Reka Edge 2603$0.10$0.10—ChatMulti-turn conversations via /v1/chat/completions.StreamingTokens arrive as they are generated with stream: true.Image inputAccepts images alongside text in the prompt.Tool callingNative tool calls, including parallel calls.+1Structured outputsJSON Schema responses; JSON mode on most models.
reka-flash-3Reka Flash 3$0.10$0.20—ChatMulti-turn conversations via /v1/chat/completions.StreamingTokens arrive as they are generated with stream: true.ReasoningThinks before answering; the trace is billed as output.Structured outputsJSON Schema responses; JSON mode on most models.
Chat
Streaming
Image input
Tool calling
Reasoning
Structured outputs
Logprobs
Prompt caching

Note All rates are USD per million tokens. Cached input applies automatically when a request repeats a prompt prefix; see Prompt caching. On reasoning models the thinking trace is billed as output tokens. Hover a feature icon for details, or see Models for context windows and output limits.

Next steps