[DOCS]

Chat completions

POST /v1/chat/completions accepts the standard OpenAI request and response shapes, streaming included. Unknown parameters are silently ignored.

REQUEST

POST https://blazeinference.com/v1/chat/completions

{
  "model": "deepseek-ai/deepseek-v4-pro",
  "messages": [
    { "role": "system", "content": "You are terse." },
    { "role": "user", "content": "Summarize SSE in one line." }
  ],
  "temperature": 0.7,
  "max_tokens": 400
}

Parameters

PARAM	TYPE	NOTES
`model`	string	Required. An available model id; see Models.
`messages`	array	Required. OpenAI message objects (system/user/assistant/tool/developer).
`stream`	boolean	SSE streaming. Terminates with data: [DONE].
`max_tokens`	integer	Output tokens only. Capped at 2,000,000.
`max_completion_tokens`	integer	Alias for max_tokens; same cap.
`temperature`	number	Passed through.
`top_p`	number	Passed through.
`stop`	string \| string[]	Passed through.
`seed`	integer	Passed through.
`frequency_penalty / presence_penalty`	number	Passed through.
`n`	integer	Must be 1.
`tools / tool_choice`	array / any	Function calling, passed through.
`response_format`	object	e.g. JSON mode, passed through.
`logprobs / top_logprobs`	boolean / integer	Passed through.
`user`	string	End-user identifier, passed through.
`stream_options`	object	e.g. {"include_usage": true} for usage in streams.
`chat_template_kwargs`	object	Advanced, optional; see below.

Streaming

Set stream: true to receive server-sent events: data: chunks terminated by data: [DONE], exactly like OpenAI. If generation fails mid-stream, you receive one final event carrying our standard error envelope, then [DONE]; a request that fails this way is not counted against your quota.

STREAMING · openai sdk

const stream = await blaze.chat.completions.create({
  model: "z-ai/glm-5.2",
  messages: [{ role: "user", content: "Write a haiku about GPUs." }],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content || "");
}

Reasoning (DeepSeek)

deepseek-ai/deepseek-v4-pro and deepseek-ai/deepseek-v4-flash support on-demand reasoning. By default we run them with reasoning off for predictable latency; opt in per request with chat_template_kwargs, and your value always overrides the default. Flash additionally accepts "reasoning_effort": "low" | "medium" | "high". When reasoning is on, the assistant message may carry a reasoning_content field alongside content.

REASONING ON

{
  "model": "deepseek-ai/deepseek-v4-pro",
  "messages": [{ "role": "user", "content": "Prove it step by step." }],
  "chat_template_kwargs": { "thinking": true }
}

Limits

LIMIT	VALUE
Free requests / day	100; only successful (200) responses count
Request body	5 MB
max_tokens ceiling	2,000,000 (output tokens)
n	1