Chat Completions API

Generate responses to text prompts using the standard chat completion format.

Endpoint

POSThttps://api.oxlo.ai/v1/chat/completions

Parameters

Name	Type	Required	Description
model	string	Yes	ID of the model to use (e.g., `mistral-7b`, `llama-3-8b`).
messages	array	Yes	A list of messages comprising the conversation so far.
max_tokens	integer	No	Maximum number of tokens to generate. Defaults to 256.
temperature	float	No	Sampling temperature between 0 and 2. Defaults to 0.7.
stream	boolean	No	Whether to stream back partial progress. Defaults to `false`.
tools	array	No	A list of tools the model may call. Each tool is a function with a name, description, and JSON Schema parameters. See Tool Calling.
tool_choice	string	No	Controls which tool is called. `"auto"` (default) lets the model decide, `"none"` disables tool calling.
include_reasoning	boolean	No	Whether to include `reasoning_content` in responses from thinking models. Defaults to `true`. Set to `false` for cleaner output. See Reasoning.

Full Reference: For the complete list of parameters including top_p, frequency_penalty, top_k,repeat_penalty, min_p, Mirostat, and more — see the Parameters Reference.

Example Request

bash

curl https://api.oxlo.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "mistral-7b",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Hello!"
      }
    ]
  }'

Example Response

json

{
  "id": "chatcmpl-123",
  "object": "chat.completion",
  "created": 1677652288,
  "model": "mistral-7b",
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "content": "Hello there, how may I assist you today?"
    },
    "finish_reason": "stop"
  }],
  "usage": {
    "prompt_tokens": 9,
    "completion_tokens": 12,
    "total_tokens": 21
  }
}

Tool Calling

Oxlo supports OpenAI-compatible tool calling (function calling) across 15+ models. Define functions as tools, and the model will return structured tool_calls when it decides to invoke them.

Supported Models: deepseek-v3-0324, deepseek-v3.2, deepseek-v4-flash, llama-4-maverick-17b, qwen-3-32b, qwen-3-coder-30b, gpt-oss-120b, gpt-oss-20b, kimi-k2.5, kimi-k2.6, kimi-k2-thinking, oxlo-claw, mistral-7b, glm-5, minimax-m2.5

Tool Calling Request

bash

curl https://api.oxlo.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "deepseek-v3.2",
    "messages": [
      {"role": "user", "content": "What is the weather in Tokyo?"}
    ],
    "tools": [
      {
        "type": "function",
        "function": {
          "name": "get_weather",
          "description": "Get the current weather in a given location",
          "parameters": {
            "type": "object",
            "properties": {
              "location": {"type": "string", "description": "City name"}
            },
            "required": ["location"]
          }
        }
      }
    ],
    "tool_choice": "auto"
  }'

Tool Calling Response

When the model invokes a tool, the response includes a tool_calls array and finish_reason is set to "tool_calls".

json

{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "model": "deepseek-v3.2",
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "content": null,
      "tool_calls": [{
        "id": "call_abc123",
        "type": "function",
        "function": {
          "name": "get_weather",
          "arguments": "{\"location\": \"Tokyo\"}"
        }
      }]
    },
    "finish_reason": "tool_calls"
  }]
}

Parallel Tool Calling

Most supported models can invoke multiple tools in a single response. Simply define multiple functions in the tools array and the model will return multiple entries in tool_calls when needed.

⚠️ Note: Some models (e.g., llama-3.3-70b) only support a single tool definition per request. Oxlo automatically handles this — if a model rejects multiple tools, we retry with a single tool.

Reasoning Models

Some models include a reasoning_content field in their responses, which contains the model's internal thinking process. This is useful for debugging and understanding how the model arrived at its answer.

Reasoning Models: deepseek-r1-70b, deepseek-r1-8b, kimi-k2.6, kimi-k2.5, gpt-oss-120b, qwen-3-32b (when using thinking mode), qwen-3-coder-30b, minimax-m2.5

Controlling Reasoning Output

Set include_reasoning to false to omit the thinking content and receive only the final answer — similar to how other inference providers handle it.

bash

# With reasoning (default)
curl https://api.oxlo.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-r1-70b",
    "messages": [{"role": "user", "content": "What is 25 * 37?"}],
    "include_reasoning": true
  }'

# Without reasoning (cleaner output)
curl https://api.oxlo.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-r1-70b",
    "messages": [{"role": "user", "content": "What is 25 * 37?"}],
    "include_reasoning": false
  }'

Reasoning Response

When include_reasoning is true (default), the response includes both the thinking process and the final answer.

json

{
  "choices": [{
    "message": {
      "role": "assistant",
      "content": "25 × 37 = 925",
      "reasoning_content": "Let me calculate 25 * 37.\n25 * 37 = 25 * (40 - 3) = 1000 - 75 = 925"
    },
    "finish_reason": "stop"
  }]
}

Tip: Oxlo automatically extracts <think> tags from models that use them (DeepSeek R1, Qwen 3) and normalizes them into the reasoning_content field, so you always get a consistent API format regardless of the underlying model.

Error Handling

Code	Description
401	Unauthorized. Invalid or missing API key.
403	Forbidden. Access denied (e.g., plan limit reached, or model requires upgrade).
429	Too Many Requests. Rate limit exceeded.
502	Bad Gateway. Worker unreachable or returned an invalid response.
503	Service Unavailable. All workers busy or queue full.
504	Gateway Timeout. Model took too long to generate a response.