The OpenAI-compatible chat endpoint. Text models from GPT, Claude, Gemini, DeepSeek, Qwen, Kimi, GLM, Grok and others can all be called through it. To switch models, change model.
POST https://studio.dataeyes.ai/v1/chat/completions
{"id":"chatcmpl-xxx","object":"chat.completion","created":1792856685,"model":"gpt-6-luna","choices":[{"index":0,"message":{"role":"assistant","content":"I’m ChatGPT, an AI assistant here to help with questions, ideas, and tasks."},"finish_reason":"stop"}],"usage":{"prompt_tokens":22,"completion_tokens":22,"total_tokens":44,"prompt_tokens_details":{"cached_tokens":0},"completion_tokens_details":{"reasoning_tokens":0}}}
Add an image_url part to the content array of a user message. Both public image URLs and Base64 data URLs are supported.
Image URL
Local image (Base64)
The description is returned in choices[0].message.content.
Image URL download failed
With a URL, the model service downloads the image. If you get invalid_image_url (download timed out or the URL is unreachable), send the image as Base64 or use a faster image host.
Tool calling takes two steps: the model first returns the function to call and its arguments; you run it locally and send the result back; the model then writes the final answer.
GPT-6 models with tools on this endpoint
To use tools with GPT-6 models on Chat Completions, you must set "reasoning_effort": "none". Otherwise you get a 400 error: Function tools with reasoning_effort are not supported. To use reasoning and tools together, use the Responses API. Other models (Claude, DeepSeek, etc.) do not have this restriction.
Step 1: send the request with tool definitions
The model returns tool_calls, and finish_reason is tool_calls:
GPT reasoning models use reasoning_effort to control how much they think. Higher effort gives more thorough answers but takes longer and uses more tokens. Reasoning tokens are reported in usage.completion_tokens_details.reasoning_tokens and billed as output.
DeepSeek and some other models also return their reasoning in message.reasoning_content, with the final answer in content.
Claude, Gemini and other models also work on this endpoint with the same request format:
For Claude-native features such as extended thinking and prompt caching, use the Messages API. For Gemini search, code execution and other native tools, see the Gemini API.