Xiaomi model for text, images, video and audio understanding. Provider-service retirement is scheduled for October 21, 2026; app availability is separate.
Tool support requires the appropriate API integration; a supported tool is not automatically active in every chat.
Supported API features and tools
Multimodal understanding across text, images, video and audio
Published 1M context and 128K maximum output
Deep thinking, streaming and tool calls documented by Xiaomi
Structured output, context caching and web search in supported integrations
Before you choose
Multimodal understanding produces text. Audio input is not speech synthesis, and video input is not video generation.
The published K/M labels are retained rather than converted into invented exact token counts.
The reviewed sources do not establish a knowledge cutoff. Supply current evidence for time-sensitive questions; a large context window is not live web access.
Provider retirement affects migration planning even if an old catalog URL remains available.
Extended thinking
Xiaomi documents deep thinking and shows a thinking.type disabled request example. This guide does not assert an unverified default, budget or low/medium/high effort mapping. Check the current endpoint and app controls.
The provider documents compatible API request formats, but model-specific media and thinking fields must follow Xiaomi's contract.
A system message in a sample request is not evidence of a model's actual training cutoff. No cutoff is claimed here.
Test tool results, output formats and failure behavior on the migration target. The V2.6 recommendation does not automatically choose or enable a replacement in your app.
Put it to work
Start with a more useful prompt.
Original examples from EZ Ai Assist. Adapt these to your task and the features available in your workspace.
Workflow 01
Reconcile meeting media
Keep conclusions tied to the supplied recording and slides.
Compare this recording with the supplied slide images and transcript. Produce a decision log with decision, evidence timestamp or slide, owner if stated and unresolved question. Flag disagreements between sources. Do not infer approvals, people or events that are not present in the supplied material.
Workflow 02
Inspect an instructional video
Separate visible actions from unverified outcomes.
Review this tutorial video and accompanying instructions. List each visible step with a timestamp, note any skipped or unclear step, and propose a small verification test. Distinguish what is seen or heard from inferred behavior. Do not claim that an action succeeded merely because the narrator says it did.
Workflow 03
Build a multimodal migration test
Compare replacements using the same evidence.
Create an evaluation plan for migrating this multimodal workflow. Use the supplied examples to define tests for image details, audio evidence, video timing and text synthesis. Specify a scoring rubric, failure categories and human-review thresholds. Leave model results and latency blank until measured.
Developer reference
Xiaomi MiMo API pricing
These are provider API reference prices, not EZ Ai Assist subscription prices.
Xiaomi MiMo published USD API rates · USD per 1,000,000 tokens
Token type
Price
Input · cache miss
$0.14
Input · cache hit
$0.0028
Output
$0.28
These are the current model page's USD rates, not currency conversions from launch pricing or Token Plan subscription allowances.
Cache-hit rates apply only to eligible cached input, not all prompts or output. Recheck current pricing and lifecycle status before calling the API.
Common questions
A few things worth knowing.
When is the provider service scheduled to retire?
October 21, 2026, at 10:00 Beijing time, according to Xiaomi's subscription documentation. This is a provider-service notice, not a statement of EZ Ai Assist availability.
Does Pro accept the same media as the base model?
No. The Pro page documents text input; the base MiMo-V2.5 page documents text, images, video and audio. Both return text.
Which replacement should I use?
Xiaomi points toward the V2.6 series. Verify the exact available endpoint and your modality, tool and output requirements before selecting a replacement; this guide does not invent a local replacement route.
Why is the knowledge cutoff not verified?
The reviewed model pages do not establish it. A cutoff string inside a sample system prompt is caller-supplied text, not authoritative model metadata.
Does this page guarantee access in EZ Ai Assist?
No. A catalog entry and provider capability reference do not establish current app access, enabled tools or available settings. Check the model selector and controls in your workspace.
Are these EZ Ai Assist subscription prices?
No. Provider API rates are developer references. Your subscription and any limits shown in the app are separate; use the website pricing page for plans.
How should I verify an answer?
Provide the relevant evidence and acceptance criteria, ask for source references or reproducible checks, and review the result. Distinguish a proposed tool action from a tool action actually executed.
Check the source
Official documentation
Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.