Skip to main content

Chat Completions API

The OpenAI-compatible /v1/chat/completions endpoint is the primary model invocation entry point on the Turing Platform. Models from all major providers—OpenAI, Claude, Gemini, Qwen, DeepSeek, Zhipu, ByteDance, KIMI, MiniMax, Bedrock, and more—can be called through this single protocol by simply switching the model field.

Full API Reference + Try-It

Parameter schema, response schema, multi-language code samples, and an interactive Try-It console: Create chat completion →

This page focuses on usage scenarios and cross-provider capability differences; it does not repeat the schema field dictionary.

Overview​

ItemDetails
EndpointPOST /api/v1/chat/completions
Protocol compatibilityOpenAI Chat Completions API
Primary usesChat / function calling / tool use / multimodal input / streaming response / reasoning mode / web search
Supported modelsRefer to the Model Pricing List; available model IDs are listed by provider
BillingReal-time token-based metering; see Billing & Usage

When to use this endpoint​

  • Unified cross-provider entry point: When a single codebase needs to switch between Claude, GPT, and Gemini, this is the preferred choice.
  • OpenAI ecosystem tooling: LangChain, LlamaIndex, Autogen, and similar frameworks default to /v1/chat/completions.
  • Existing OpenAI SDK code: Only two lines—api_key and base_url—need to change.

Other options:

  • Claude-specific capabilities (cache_control fine-grained caching, web_search_20250305 native tool, thinking block structure) → Anthropic Messages API
  • GPT-5 stateful reasoning / MCP connector → OpenAI Response API

Capability integrations​

The sections below demonstrate concepts only—i.e., what each capability looks like in a request. The complete request body schema, response schema, and multi-language samples are available at /api/create-chat-completion.

Vision (image input)​

Multimodal models (GPT-4o / Claude / Gemini / Qwen-VL, etc.) accept image_url or base64 image content:

curl $TURING_BASE_URL/chat/completions \
  -H "Authorization: Bearer $TURING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "turing/gpt-5.5",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "Describe this image"},
        {"type": "image_url", "image_url": {"url": "https://.../image.jpg"}}
      ]
    }]
  }'

Reasoning mode​

  • OpenAI / Azure reasoning models (GPT-5, o1/o3/o4): Use v1/responses for controllable reasoning.
  • Claude 4.6+ / 5.x: Use extra_body={"thinking": {"type": "adaptive"}, "output_config": {"effort": "high"}}; Opus 4.7/4.8 and Sonnet 5 do not support enabled / budget_tokens.
  • Gemini: Use extra_body={"thinkingConfig": {"includeThoughts": True}} to surface a reasoning summary; omitting thinkingLevel / thinkingBudget uses the model's default configuration.
  • Qwen thinking variants / DeepSeek R1: Reasoning is enabled automatically; no flag required.

See Reasoning Model Configuration for details.

Pass web search configuration via tools / extra_body according to each provider's protocol. The syntax differs across providers:

  • Qwen: extra_body={"enable_search": True}
  • Gemini: tools: [{"googleSearch": {}}]
  • Claude: tools: [{"type": "web_search_20250305", "name": "web_search", "max_uses": 3}]

See Model Built-in Web Search for details.

Prompt caching​

  • Claude: Requires explicit cache_control; see Prompt Caching (Claude).
  • OpenAI GPT-4o / GPT-5: Cached automatically; no parameters needed.
  • Cache hit counts are reported in usage.prompt_tokens_details.cached_tokens and billed at the corresponding cache-read unit price.

Provider-specific parameters​

Use the OpenAI SDK's extra_body to pass any provider-specific, non-standard fields:

curl $TURING_BASE_URL/chat/completions \
  -H "Authorization: Bearer $TURING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-max-latest",
    "messages": [...],
    "enable_search": true,
    "thinking": {"type": "adaptive"},
    "cache_control": {"type": "ephemeral"}
  }'

With cURL, write these fields directly at the top level of the request body. With the Python SDK, use extra_body={...}; with the Java SDK, use putAdditionalBodyProperty; with the TypeScript SDK, use @ts-expect-error or as any to bypass type checking.

turing_options​

The Turing Platform extends the OpenAI request body with a turing_options field (placed at the top level of the request body, not inside extra_body) for configuring platform behaviors such as timeouts, retries, and fallback. For field definitions, default values, and allowed ranges, see Timeouts, Retries & Fallback.

See also​