Chat Completions

POST https://api.inferam.ai/v1/chat/completions is OpenAI’s Chat Completions format. It’s the model server’s native format, so requests pass through with the fewest changes.

curl https://api.inferam.ai/v1/chat/completions \
  -H "Authorization: Bearer $INFERAM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "DeepSeek-V4.1-Flash-1017",
    "stream": true,
    "stream_options": {"include_usage": true},
    "reasoning_effort": "low",
    "messages": [
      {"role": "system", "content": "You are a terse senior engineer."},
      {"role": "user", "content": "Why is my Go map iteration order random?"}
    ]
  }'

Supported fields

model, messages, stream, stream_options.include_usage, max_tokens, max_completion_tokens, temperature, top_p, stop, seed, presence_penalty, frequency_penalty, tools, tool_choice, parallel_tool_calls, response_format (text, json_object, json_schema), reasoning_effort and thinking_budget.

Other fields (n, logprobs, best_of, logit_bias, engine-specific sampling options, and so on) are ignored.

• Roles: system, developer, user, assistant, tool.
• Tools: function tools with JSON Schema parameters, parallel tool calls and tool_choice (auto, none, required or a named function).
• Images: image_url content parts with a base64 data: URL. Remote URLs are rejected (see The model).

Reasoning

The model thinks before it answers by default. Its reasoning arrives in message.reasoning_content, or in delta.reasoning_content when streaming, separate from content. When you send an assistant turn that made tool calls back to the model, include its reasoning_content: the model expects to see its earlier reasoning on tool-calling turns.

• reasoning_effort: none (thinking off), minimal, low, medium, high, xhigh or max.
• thinking_budget: a cap on reasoning tokens, no larger than the output limit.

Streaming

With "stream": true the response is server-sent events of chat.completion.chunk objects, ending with data: [DONE]. Set "stream_options": {"include_usage": true} to get a final chunk with usage and no choices.

Usage

"usage": {
  "prompt_tokens": 2048,
  "completion_tokens": 312,
  "total_tokens": 2360,
  "prompt_tokens_details": { "cached_tokens": 1792 }
}