MiMo parameter and protocol differences

For MiMo models, adjust your chosen protocol's parameters using this page. See Chat Completions, Responses and Messages for full request and response structures.

Applicable models

Model	Applies to
mimo-v2.6-pro-ultraspeed, mimo-v2.6-flash, mimo-v2.6-pro	All support Chat, Responses and Messages, plus image, audio and video input; the following rules apply to all three models

Chat Completions differences

Field	Difference
`thinking.type`	Supports `enabled/disabled`, defaulting to `enabled`; enables or disables thinking
`temperature`, `top_p`	Ranges: `0～1.5` and `0.01～1`; defaults: `1.0` and `0.95`, respectively. Thinking mode overrides supplied values with these defaults; non-thinking mode allows adjustment
`max_completion_tokens`	Range: `1～131072`; default: `131072`; thinking and the final answer share this budget
`stop`	At most `4` stop strings
`response_format.type`	The vendor lists `text/json_object`; JSON mode also requires explicitly requesting JSON in the prompt
`tool_choice`	The effective strategy is `auto`; other values are filtered and treated as `auto`, so they neither force nor prevent tool calls
Historical `reasoning_content`	In multi-turn thinking-mode tool conversations, preserve complete historical assistant messages and thinking; missing thinking for tool turns returns `400`
`web_search` in `tools`	MiMo search extension with options such as `force_search`, `max_keyword` and `limit`; sources appear in `message.annotations`, and call counts in `usage.web_search_usage`
`finish_reason`	Also includes `repetition_truncation`, indicating repetitive generation was detected and truncated

Responses differences

Field	Difference
`reasoning.effort`	Thinking is enabled by default; `none` disables it and all `minimal/low/medium/high/xhigh/max/ultra` enable it, with no difference in actual reasoning depth currently
`temperature`, `top_p`	Ranges and thinking-mode overrides follow the Chat table
`max_output_tokens`	Range: `1～131072`; default: `131072`; includes thinking and the final answer
`text.format.type`	The vendor lists `text/json_object`; for JSON mode, specify the output structure in the prompt
`tool_choice`	The effective strategy is `auto`; other values are filtered and treated as `auto`
`tools[].type`	The vendor lists `function/custom/namespace`, supporting function tools, custom tools and tool namespaces
`previous_response_id`, `background`, `context_management`	Unsupported; omit these fields. Replay historical input and output on the client for multi-turn conversations
Historical reasoning items	In multi-turn tool conversations, retain the returned reasoning and tool-call items in subsequent `input`

Messages differences

Field	Difference
`max_tokens`	Optional; range: `1～131072`; default: `131072`; includes thinking and the final answer
`thinking.type`	Supports `enabled/disabled`, defaulting to `enabled`; use this switch to control thinking
`temperature`, `top_p`	Ranges and thinking-mode overrides follow the Chat table
`messages[].role`	Allows `system` in addition to `user/assistant`
`tool_choice.type`	The effective strategy is `auto`; other values are filtered and treated as `auto`
`tool_choice.disable_parallel_tool_use`	When `true`, returns at most one tool call under strategy `auto`
Historical thinking blocks	Preserve complete assistant content, including thinking, text and tool_use blocks, and replay it with tool results
`stop_reason`	Also includes `content_filter` and `repetition_truncation`, indicating content filtering and repetitive-generation truncation, respectively

Cache differences

Field	Difference
Context caching	All three models support caching; inspect response usage fields to confirm actual hits
Chat	`usage.prompt_tokens_details.cached_tokens` reports cached input tokens
Responses	`usage.input_tokens_details.cached_tokens` reports cached input tokens
Messages	`usage.cache_read_input_tokens` reports input tokens read from cache

Media input differences

Input	Protocol content block	Difference
Image	`image_url`	Chat: pass a public URL or Base64 Data URL in `image_url.url`
Audio	`input_audio`	Chat: pass a public URL or Base64 Data URL in `input_audio.data`
Video	`video_url`	Chat: pass a public URL or Base64 Data URL in `video_url.url`; put `fps` and `media_resolution` on the content block, alongside `video_url`
Image	`input_image`	Responses: `image_url` is a public URL or Base64 Data URL string
Audio	`input_audio`	Responses: uses a `audio_url` string, unlike Chat's `input_audio.data`
Video	`input_video`	Responses: uses a `video_url` string; `fps` and `media_resolution` belong on the same content block
Image	`image`	Messages: provide `url` for `source.type=url`; provide `media_type` and raw Base64 `data` for `source.type=base64`
Audio	`audio`	Messages extension: pass a URL or Base64 through `source`, with the same fields as image sources
Video	`video`	Messages extension: pass a URL or Base64 through `source`; `fps` and `media_resolution` belong inside `source`

Field	Difference
`fps`	Range: `0.1～10`; default: `2`; controls sampled video frames per second
`media_resolution`	Supports `default/max`; default: `default`; controls video frame resolution
Image size	URL files: at most `50 MB`; each image's Base64-encoded string: at most `50 MB`
Audio size	URL files: at most `100 MB`; each audio clip's Base64-encoded string: at most `50 MB`
Video size	URL files: at most `300 MB`; each video's Base64-encoded string: at most `50 MB`

Request examples

Replace MODEL_ID with the full platform model ID and use your own API Key.

curl 'https://api.agtcloud.ai/v1/chat/completions' \
  -H 'Authorization: Bearer your-api-key' \
  -H 'Content-Type: application/json' \
  --data '{
    "model": "MODEL_ID",
    "messages": [{"role": "user", "content": "Explain how to troubleshoot an API timeout in three points"}],
    "thinking": {"type": "enabled"},
    "max_completion_tokens": 4096
  }'

Official references

Pro UltraSpeed · Flash · Pro
Chat parameters and responses · Responses parameters and responses · Messages parameters and responses
Model hyperparameters · Deep thinking and history replay
JSON output · Web search
Image input · Audio input · Video input

MiMo parameter and protocol differences

On this page