---
title: "Request parameters"
description: "Set output limits, sampling controls and tool calls for your API requests."
canonical_url: "https://minirouter.sh/docs/parameters"
markdown_url: "https://minirouter.sh/docs/parameters.md"
last_updated: "2026-09-13"
---

# Request parameters

Control the answer length, sampling and tool use. The examples below use Chat
Completions; other endpoints use different field names.

## Set your key

Create an [API key](https://minirouter.sh/key), save its recovery link, and
[add credits](https://minirouter.sh/dashboard/billing). Replace the placeholder
below and run it in your terminal before running the examples on this page.

```sh
export MINIROUTER_KEY='paste-your-api-key-here'
```

Examples use paid models. For free requests, follow the
[Auto Free guide](https://minirouter.sh/docs/free-inference).

## Send a request with parameters

Run this in the same terminal. Change the prompt, temperature or output limit
to suit your request.

```sh
curl https://api.minirouter.sh/v1/chat/completions \
  -H "Authorization: Bearer $MINIROUTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "openai/gpt-4.1-mini",
  "messages": [
    {
      "role": "system",
      "content": "Give short, practical answers."
    },
    {
      "role": "user",
      "content": "Give three names for a gardening app."
    }
  ],
  "max_tokens": 256,
  "temperature": 0.7
}'
```

Read the answer in `choices[0].message.content`. A `finish_reason` of
`length` means the output limit was reached; increase it for a longer answer.

## Common Chat Completions fields

| Field | How to use it |
| --- | --- |
| `model` | Required. Copy an exact model ID or listed alias from the [catalog](https://minirouter.sh/models). |
| `messages` | Required, non-empty conversation array. Use `system` for instructions, `user` for prompts and `assistant` for earlier answers. Send the history needed for each request. |
| `max_tokens` | Positive integer limiting generated output. Keep it within the model limit. |
| `max_completion_tokens` | Alternative output-limit field used by some models. Send this or `max_tokens`, never both. |
| `temperature` | Controls sampling randomness. Lower values generally make answers more focused. Accepted ranges depend on the model. |
| `top_p` | Nucleus sampling. Tune this or temperature first, rather than changing both together. |
| `stop` | A string or list of stop sequences. Output ends when a supported model reaches one. |
| `frequency_penalty`, `presence_penalty` | Adjust repetition on models that support these controls. |
| `stream` | Set to `true` for server-sent events. See [streaming](https://minirouter.sh/docs/streaming). |
| `stream_options` | Chat Completions accepts `{"include_usage":true}` for streaming usage. |
| `reasoning`, `reasoning_effort` | Thinking controls. Choose one form; see [reasoning](https://minirouter.sh/docs/reasoning). |
| `tools`, `tool_choice` | Describe functions your application can run and control when the model calls them. Example below. |
| `parallel_tool_calls` | Boolean controlling parallel calls on models that support them. |
| `seed` | Integer sampling seed where supported. It does not guarantee identical answers. |
| `n` | Only `1` is supported. |
| `logprobs`, `top_logprobs` | Request token probabilities where supported. `top_logprobs` accepts an integer from 0 to 20. |
| `models` | Ordered text-model fallback list. See [fallback models](https://minirouter.sh/docs/models). |

Optional parameters vary by model and provider. Check `supported_parameters`
on the [public catalog](https://minirouter.sh/api/v1/models), plus
`reasoning_options` for thinking controls. If a field is rejected, read the
field path in the error and remove or adjust it. For schema-constrained JSON,
use the [structured-output guide](https://minirouter.sh/docs/structured-output).

## Field names by endpoint

| Setting | Chat Completions | Responses | Messages |
| --- | --- | --- | --- |
| Endpoint | `/v1/chat/completions` | `/v1/responses` | `/v1/messages` |
| Input | `messages` | `input` | `messages` |
| System instructions | A system message | `instructions` | `system` |
| Output limit | `max_tokens` or `max_completion_tokens` | `max_output_tokens` | `max_tokens` (required) |
| Thinking | `reasoning` or `reasoning_effort` | `reasoning` | `thinking` and `output_config.effort` |
| Schema output | Not supported | Not supported | `output_config.format` on the models listed in the structured-output guide |

Use the request shape for your endpoint. Moving the same JSON body between
endpoints without changing its fields will return an error.

## Call a tool

Send a function definition with a model that supports tools. MiniRouter returns
the model's requested call; your application runs the function.

```sh
curl https://api.minirouter.sh/v1/chat/completions \
  -H "Authorization: Bearer $MINIROUTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "openai/gpt-4.1-mini",
  "messages": [
    {
      "role": "user",
      "content": "What is the weather in London?"
    }
  ],
  "max_tokens": 256,
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "get_weather",
        "description": "Get the current weather for a city.",
        "parameters": {
          "type": "object",
          "properties": {
            "city": {
              "type": "string"
            }
          },
          "required": [
            "city"
          ],
          "additionalProperties": false
        }
      }
    }
  ],
  "tool_choice": "auto"
}'
```

1. Read `choices[0].message.tool_calls`. With `tool_choice: "auto"`, the model may answer directly instead.
2. Parse each function's `arguments` JSON and validate it before running your function.
3. Append the complete assistant message, including `tool_calls`, to the conversation.
4. Append a `role: "tool"` message for each result, using the matching `tool_call_id` and a string `content`.
5. Send the updated conversation to get the final answer.

For example, a tool result message looks like
`{"role":"tool","tool_call_id":"call_id_from_response","content":"18 C, cloudy"}`.
Use the ID from the actual response. To force a call, use `tool_choice: "required"`;
to disable calls, use `"none"`. These fields require a model that supports tool choice.

## Handle an error

A `400 invalid_request` names the field to change. Reduce oversized schemas,
remove unsupported controls, and check that you used the right endpoint.
A `402 insufficient_credits` means the available balance cannot cover the request;
reduce the output limit or add credits. See the [error reference](https://minirouter.sh/docs/errors).