xiaomi
Xiaomi: MiMo-V2.5
xiaomi/mimo-v2.5
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
xiaomi/mimo-v2.5 is Xiaomi's native omnimodal model: a sparse mixture-of-experts system with 310B total and 15B activated parameters, accepting text, image, video, and audio inputs and producing text, according to its Hugging Face model card and the OpenRouter catalog. The card reports a hybrid sliding-window and global attention design, dedicated vision and audio encoders, multi-token prediction for faster decoding, and a context window of up to 1M tokens. Weights are published under the MIT license with documented SGLang and vLLM deployment paths. Claimed agentic and multimodal strengths are vendor- and catalog-reported; independent evaluation evidence was limited in this source pack.[1][2]
Strengths
- Native omnimodal input handling: the model card describes unified understanding of text, images, video, and audio via a 729M-parameter vision transformer and an audio encoder initialized from MiMo-Audio, matching the modalities listed in the OpenRouter catalog.[1][2]
- Efficiency-oriented architecture for long contexts: hybrid attention reported to cut KV-cache storage by nearly 6x, multi-token prediction for speculative decoding, FP8 training, and a claimed context window of up to 1M tokens.[2]
- Open distribution and deployment support: MIT-licensed weights on Hugging Face, plus officially referenced SGLang and vLLM deployment guides; the catalog also lists tool calling, structured outputs, and optional reasoning among supported API parameters.[1][2]
Limitations
- Evidence gap: the model card's benchmark results are embedded as images with no extractable figures in this pack, and only catalog-reported benchmark data was available; the vendor's claims of strong agentic and multimodal performance should be validated on your own workloads.[1][2]
- Output is limited to text, the knowledge cutoff is not specified in the catalog entry, and the card warns that weights downloaded before a config update need refreshed config.json and tokenizer_config.json files to avoid degraded behavior.[1][2]
- The publisher's use restrictions prohibit military applications, unlawful or harmful uses, and autonomous high-risk actions without human oversight, which may constrain some deployment scenarios.[2]
Where to use it
- Multimodal assistants or pipelines that need to analyze images, video, or audio and return structured text, suggested by the documented encoders and catalog-listed input modalities.[1][2]
- Long-document or long-session analysis that could benefit from the advertised 1M-token context window and reduced KV-cache footprint, pending validation on representative inputs.[2]
- Agentic and tool-calling applications via API, given catalog-listed support for tools, structured outputs, and reasoning parameters alongside vendor-reported agentic post-training.[1][2]
Sources
- [1] OpenRouter catalog · Xiaomi: MiMo-V2.5Sources checked: Nov 13, 2026
- [2] Model documentation · Xiaomi: MiMo-V2.5Sources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
MiMo-V2.5 is a native omnimodal model by Xiaomi. It delivers Pro-level agentic performance at roughly half the inference cost, while surpassing MiMo-V2-Omni in multimodal perception across image and video understanding...
Specifications
- Model ID
- xiaomi/mimo-v2.5
- Canonical ID
- xiaomi/mimo-v2.5-20260422
- Creator
- xiaomi
- Added to OpenRouter
- 22.04.2026
- Context length
- 1,000,000
- Modality
- text+image+audio+video->text
- Input modalities
- text, audio, image, video
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.400 / out $2
- Knowledge cutoff
- —
- Expiration date
- —
Tasks and capabilities
A grade appears only where the source supplies a matching measurement for that task, the model is active and the cohort holds at least 10 comparable models. The capability column lists facts the source declares — supported formats, reasoning and tool parameters — which are not measured quality.
| Task | Grade | Source-declared capability |
|---|---|---|
| General quality | No measurement | reasoning: optional · parameters: include_reasoning, reasoning |
| Coding | BB — next 30% of the cohort | — |
| Agents | BB — next 30% of the cohort | parameters: structured_outputs, tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | reasoning: optional · parameters: include_reasoning, reasoning |
| Document analysis | No measurement | — |
| Vision | No measurement | input: image |
| Image generation | No measurement | — |
| Audio | No measurement | input: audio |
| Video | No measurement | input: video |
| Embeddings | No measurement | — |
Evaluations
| Source | Benchmark | Value | Grade | Data retrieved |
|---|---|---|---|---|
| Artificial Analysis | Coding index | 56.8 | BB — next 30% of the cohort | 13.11.2026 |
| Artificial Analysis | Agentic index | 15.8 | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · 3dwin rate 50.9% · rank 49 | 1,200Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · asciiartwin rate 45.1% · rank 50 | 1,139Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · codecategorieswin rate 53.7% · rank 42 | 1,253Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · datavizwin rate 53.9% · rank 35 | 1,250Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · gamedevwin rate 52.6% · rank 41 | 1,243Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · svgwin rate 50.3% · rank 39 | 1,176Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · uicomponentwin rate 53.7% · rank 44 | 1,250Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · websitewin rate 53.9% · rank 36 | 1,261Elo | AA — next 20% of the cohort | 13.11.2026 |
How grades work
Each letter grade compares a model only against other models scored by the same source and metric in the current snapshot. S means the top 10%, A the next 20%, B the next 30%, C the rest. Models without a published measurement are shown as not rated — a supported format alone never implies measured quality.
- S — S — top 10% of the cohort
- A — A — next 20% of the cohort
- B — B — next 30% of the cohort
- C — C — below the B band
Providers
Optional live provider statistics from OpenRouter. Missing values are unknown, not zero.
| Provider | Context length | Price in /1M | Price out /1M | Latency p50 (30m, s) | Throughput p50 (30m, tok/s) | Uptime (30m) |
|---|---|---|---|---|---|---|
| Venicefp8 | 1,000,000 | $0.400 | $2 | — | — | 100% |
Provider data retrieved: 13.11.2026
Related articles
No published articles identify this model yet.