Xiaomi MiMo
Xiaomi's open MIT-licensed model family — near-frontier at low cost
Overview
MiMo is Xiaomi's open-weight model family spanning reasoning, coding and multimodal generation. The MiMo-V2.5-Pro flagship tops 1 trillion parameters (42B active, 1M context) and ranks among top Chinese LLMs, while lighter Flash/V2.5 variants target agentic and coding work — all MIT-licensed.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 13 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
What you can produce with Xiaomi MiMo
- MIT-licensed open weights (Flash + V2.5)
- MiMo-V2-Flash: 309B total / 15B active, 256K context
- MiMo-V2.5-Pro: 1T+ params, 1M context, multimodal
- Top open-source scores on SWE-bench Verified/Multilingual
- First-party token-based hosted API + self-host option
- Repos and weights on GitHub (XiaomiMiMo) and Hugging Face
ASEAN Perspective
Xiaomi MiMo in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Xiaomi turned out to be a serious open-model player, not a phone-maker dabbling. The MIT license is the headline: commercial use, fine-tuning and derivative weights with no revenue caps. On capability the split matters — MiMo-V2-Flash (309B total, ~15B active) is the standout, ranking as a top open-source model on SWE-bench Verified and Multilingual at roughly Claude Sonnet 4.5 quality for about 3.5% of the cost. The larger MiMo-V2-Pro/V2.5-Pro (1T+ params, 42B active, 1M context) reaches the upper tier of Chinese LLMs and the top ten on Artificial Analysis' index, but note the Pro tier is the proprietary hosted line, not the open one. Independent aggregators are less breathless than the launch coverage — BenchLM puts Flash mid-pack overall (#55/124) while acknowledging its coding strength. Best for: developers who want cheap, permissive, self-hostable coding and agentic models and don't need a single closed frontier API.
What people say
MiMo went from a curiosity — a phone company shipping a reasoning model in 2025 — to a genuinely competitive open-weight family by mid-2026, and the coverage reflects that shift.
The clearest signal is at the efficient end. MiMo-V2-Flash, a 309B-total / 15B-active MoE with a 256K window, is described in Xiaomi's technical report and by VentureBeat as the top open-source model globally on SWE-bench Verified and SWE-bench Multilingual, landing near Claude Sonnet 4.5 quality on those coding tasks while costing roughly 3.5% as much per token. Independent aggregator BenchLM.ai is more measured: it ranks Flash #55 of 124 models overall (score 59/100) and #20 of 124 in mathematics (82.1 average), which reads as 'excellent for coding, middling as an all-rounder.'
At the top of the range, MiMo-V2-Pro and the multimodal MiMo-V2.5-Pro cross 1 trillion parameters with 42B active and up to a 1M-token context. Per the Artificial Analysis Intelligence Index cited in coverage, MiMo-V2-Pro ranks around 8th worldwide and 2nd among Chinese LLMs; VentureBeat framed V2-Pro as nearing GPT-5.2 and Opus 4.6 performance at a fraction of the cost. The V2.5-Pro variant reportedly handles text, image, audio and video in one model at 60-80 tokens/sec, with lighter builds hitting 100-150 tokens/sec.
Licensing is a real selling point rather than a footnote. MiMo-V2.5 and the code tooling ship under the MIT License, which reviewers repeatedly call out for allowing commercial deployment, private fine-tuning and derivative-weight release without permission or usage caps. Weights and code live under the XiaomiMiMo GitHub org and Hugging Face account.
Two things to keep straight. First, the naming: 'Flash' and the base V2.5 are the open, self-hostable models, while the 'Pro' tier is Xiaomi's proprietary hosted line — easy to conflate when reading benchmark charts. Second, most performance claims for the top models trace back to Xiaomi's own reports and the Artificial Analysis index rather than a broad base of enterprise G2/Gartner reviews, which barely exist for MiMo yet. For coding and agentic work on a budget the value case is strong; for mission-critical general reasoning, verify against your own tasks.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Xiaomi MiMo's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Xiaomi MiMo unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Xiaomi MiMo directly →
Spotted something out of date? Suggest an update →
Alternatives to Xiaomi MiMo
More in LLMs & Chat
