Xiaomi · Model guide
MiMo V2.6 Pro
Xiaomi's flagship MiMo V2.6 model: an open-weight sparse MoE with 1.02T total and 42B active parameters, a 1M-token context and native understanding of text, images, audio and video. Trained with a single large-scale reinforcement-learning run spanning coding, general agents, visual tasks and cybersecurity, it targets complex projects, long-horizon agent work, cybersecurity and scientific research.
What this model is good at
What the vendor positions it for, and which of our models to reach for instead.
VendorXiaomi's most powerful flagship reasoning model: omni-modal, trillion-parameter, built for complex projects, long-horizon tasks, high-stakes work, cybersecurity and research.source
Long-horizon coding agents · Scores 71.9 on DeepSWE v1.1 and 89.9 on Terminal-Bench 2.1 in Xiaomi's own evaluation, up from 19.0 and 65.2 for MiMo V2.5 Pro.(vendor claim)source
Cybersecurity and scientific research · Xiaomi's selection guide recommends it for cybersecurity and scientific research, and the release shows it co-designing materials and formalising a theorem in Lean 4.(vendor claim)source
Omni-modal agents · Takes text, images, audio and video in one model, and scores 82.0 on OSWorld-Verified for computer use.(vendor claim)source
- xiaomi/mimo-v2.6-flashsibling —Choose Flash for high-frequency calls and large-scale workloads where Pro's extra headroom is not needed.
Pricing and billing
Billed per token. Cached input is charged at the cache-read rate.
| Price / 1M tokens | You pay |
|---|---|
| Input | $ 0.47 |
| Output | $ 0.94 |
| Cache read | $ 0.0050 |
| Cache write | $ 0.47 |
+ 2.0K × $ 0.94
Capabilities and limits
What CrossModel guarantees across every route this model can take right now.
- Context window
- 1.0M tokens
- Max output
- 131.1K tokens
- Input / output modalities
- Text + Image + Audio + Video → Text
- Streaming
- Supported
- Tool calling
- Supported
- Structured output (JSON)
- Supported
- Image input
- Supported
- Thinking mode
- Supported· can be turned off
- Reasoning effort
- none · low · medium · high
- Available endpoints
- /v1/chat/completions · /v1/responses · /v1/messages
Published benchmarks
Scores the vendor reported, with the evaluation setup each one came from.
- DeepSWE 1.1resolved_rate·Xiaomi MiMo · 2026-09-22setup not fully disclosed
厂商自测（技术报告 Table 3，与模型卡同表）；未披露本模型的推理档位、harness 与采样设置。发布博文正文另给出 6 天 Live RL 结束时的数字（Pro 72.6 / Flash 65.7），与最终表不同，此处取最终表。
71.9% - ProgramBenchscore·Xiaomi MiMo · 2026-09-22setup not fully disclosed
厂商自测（技术报告 Table 3，与模型卡同表）；未披露本模型的推理档位、harness 与采样设置。
26.5%
- CyberGymscore·Xiaomi MiMo · 2026-09-22setup not fully disclosed
厂商自测（技术报告 Table 3，与模型卡同表）；未披露本模型的推理档位、harness 与采样设置。技术报告脚注：Xiaomi 按报告 4.2.4 节的方法修正了有缺陷的评测环境，与其它厂商的 CyberGym 成绩不可直接比较。
94% - ExploitBenchscore·Xiaomi MiMo · 2026-09-22setup not fully disclosed
厂商自测（技术报告 Table 3，与模型卡同表）；未披露本模型的推理档位、harness 与采样设置。
47.9%
Vendor-reported numbers, not CrossModel measurements. Scores are only comparable when the benchmark version, metric and evaluation setup match, so nothing here is averaged or ranked.
Use it in your tools
Point the base URL at CrossModel and paste this model ID — every tool below has a setup guide.
Related models
Other tiers from the same vendor, plus similarly priced alternatives.
Frequently asked questions
What is MiMo V2.6 Pro?
How much does MiMo V2.6 Pro cost?
Does MiMo V2.6 Pro support tool calling and structured output?
Which endpoint do I call?
Can I try it without writing code?
Chat first, integrate later
Chat first, then wire it in once you like the answers.