Compatibility and request fields - Mibyan Docs

Mibyan accepts the common text-generation fields used by OpenAI-compatible clients. The gateway validates and normalizes the request before it reaches the model service.

Supported request fields

Field	Chat Completions	Responses	Notes
`model`	Yes	Yes	Use `mibyan-4.1`
`messages`	Yes	Yes	Roles: `system`, `developer`, `user`, `assistant`, `tool`
`input`	No	Yes	String or message array
`instructions`	No	Yes	Converted to a developer message
`stream`	Yes	Yes	SSE output
`max_tokens`	Yes	Accepted through normalized request	Integer from 1 to 32,000
`max_output_tokens`	No	Yes	Normalized to the generation limit
`temperature`	Yes	Yes	`0` to `2`
`top_p`	Yes	Yes	`0` to `1`
`frequency_penalty`	Yes	Yes	`-2` to `2`
`presence_penalty`	Yes	Yes	`-2` to `2`
`stop`	Yes	Yes	String or a short list
`tools`	Yes, when enabled	Model-dependent	Definitions are size-limited
`tool_choice`	Yes, when enabled	Model-dependent	Passed only when tools are accepted
`response_format`	Model-dependent	Model-dependent	Requires structured-output access

Unknown fields are ignored for forward compatibility.

Message content

Text content can be a string. Structured content arrays are accepted only when the active model and endpoint support them; unsupported media or file parts should be rejected by your integration rather than assumed to work.

Not currently exposed

The public API does not currently expose embeddings, image generation, audio generation, file storage, background mode, hosted tools, or conversations. Watch the changelog for availability changes.

Stable contract

Use the documented base URL, model ID, error envelope, request ID, and standard response fields.