MiMo-V2.6-Pro is Xiaomi's open-weight AI model, released 21/09/2026. Xiaomi's model card compares MiMo-V2.6-Pro with V2.6-Flash, V2.5-Pro, Claude Opus 5, GPT-5.6 Sol and Claude Fable 5. It is level with or ahead of the closed models on AutomationBench, Terminal Bench 2.1 and Agents' Last Exam, and behind Claude Opus 5 on Terminal Bench 4.0, ProgramBench and Toolathlon. The card gives no per-benchmark settings, and rows marked internal cannot be reproduced outside Xiaomi.
Visit MiMo-V2.6-ProAI-Powered
Leverages advanced AI technology to deliver cutting-edge capabilities and results.
Fast & Efficient
Optimized performance ensures quick results without compromising on quality.
Purpose-Built
Specifically designed for llms tasks and workflows.
How does Xiaomi MiMo-V2.6-Pro score on independent benchmarks?
Xiaomi MiMo-V2.6-Pro ranks #21 of 28 models on the Artificial Analysis Intelligence Index (v5.0.1) with a score of 29.9, as measured on 22 November 2026.
Where does Xiaomi MiMo-V2.6-Pro rank against other models?
How does Xiaomi MiMo-V2.6-Pro do on individual tests?
| Evaluation (run by Artificial Analysis) | Xiaomi MiMo-V2.6-Pro | Best of any model |
|---|---|---|
| Terminal-Bench 4.0 | 34.8% | 67.2% |
| FrontierCode 2.0 | 24% | 48.8% |
| Humanity's Last Exam | 49.4% | 74.3% |
| GDPval-AA (Elo) | 1692 | 1866 |
How has Xiaomi MiMo-V2.6-Pro's score changed over time?
We have tracked Xiaomi MiMo-V2.6-Pro on Intelligence Index v5.0.1 since 22/11/2026. A point is added each day the score, coding index or price changes, so this chart fills in over time.
What does the Xiaomi MiMo-V2.6-Pro system card report?
Xiaomi's model card compares MiMo-V2.6-Pro with V2.6-Flash, V2.5-Pro, Claude Opus 5, GPT-5.6 Sol and Claude Fable 5. It is level with or ahead of the closed models on AutomationBench, Terminal Bench 2.1 and Agents' Last Exam, and behind Claude Opus 5 on Terminal Bench 4.0, ProgramBench and Toolathlon. The card gives no per-benchmark settings, and rows marked internal cannot be reproduced outside Xiaomi.
These are Xiaomi's own published results (self-reported, with the vendor's harness and settings), so they are not directly comparable with other labs' numbers. Sources: MiMo-V2.6-Pro-RL (Hugging Face model card) (22/09/2026), MiMo-V2.6 release notes (22/09/2026). Last checked 22/11/2026.
Coding
| Benchmark | Xiaomi MiMo-V2.6-Pro | MiMo-V2.6 Flash | MiMo-V2.5 Pro | Claude Opus 5 |
|---|---|---|---|---|
| DeepSWE v1.1Xiaomi harness; per-benchmark settings not published | 71.9% | 67.9% | 19% | 74% |
| ProgramBenchXiaomi harness; per-benchmark settings not published | 26.5% | 26% | 12.5% | 37% |
| MiMo Code Bench (internal)Xiaomi internal benchmark, cannot be reproduced outside Xiaomi | 63.2% | 61.2% | 40.4% | 68.6% |
| Terminal Bench 4.0 | 34.9% | 28.8% | 1.5% | 49% |
| Terminal Bench 2.1 | 89.9% | 87.6% | 65.2% | 89.1% |
| DeepSWE v1.1 (release-notes figure)Out-of-sample long-horizon SWE figure from the release notes; differs from the model card row (71.9), so run settings differ. Flash is 65.7 on the same basis | 72.6% | n/a | n/a | n/a |
Agents & tool use
| Benchmark | Xiaomi MiMo-V2.6-Pro | MiMo-V2.6 Flash | MiMo-V2.5 Pro | Claude Opus 5 |
|---|---|---|---|---|
| AutomationBench v1.0.6Xiaomi harness | 53.1% | 52.3% | 16% | 50.3% |
| Toolathlon-VerifiedXiaomi harness | 76.9% | 73.6% | 49.1% | 80.6% |
| GDPval-AA 2.1 | 1673 Elo | n/a | 1107 | 1708 |
| Agents' Last Exam | 31.6% | 27.6% | 13.2% | 31.6% |
| OSWorld-Verified | 82% | 80.8% | n/a | 83.4% |
| JobBench | 62% | 61.2% | 25% | 65.7% |
Other
| Benchmark | Xiaomi MiMo-V2.6-Pro | MiMo-V2.6 Flash | MiMo-V2.5 Pro | Claude Opus 5 |
|---|---|---|---|---|
| CyberGymCybersecurity | 94% | 95.1% | 40% | n/a |
| MiMo Cyber Bench (internal)Xiaomi internal benchmark, cannot be reproduced outside Xiaomi | 80.2% | 77.2% | 0% | n/a |
| ExploitGymCybersecurity | 17.8% | 6% | 0.2% | 22.1% |
| ExploitBenchCybersecurity | 47.9% | 25.3% | 16.6% | 70% |
| SEC Bench ProCybersecurity | 66.3% | 47.5% | 17.7% | n/a |
| Artificial Analysis Intelligence IndexThird-party index quoted in the Xiaomi release notes; Xiaomi says this is above Kimi K3 and Qwen3.8 Max | 46 index | n/a | n/a | n/a |
Multimodal
| Benchmark | Xiaomi MiMo-V2.6-Pro | MiMo-V2.6 Flash | MiMo-V2.5 Pro | Claude Opus 5 |
|---|---|---|---|---|
| MiMo VisualCoding (internal)Visual agent; Xiaomi internal benchmark | 72.3% | 71.5% | n/a | 70% |
What did safety testing find for Xiaomi MiMo-V2.6-Pro?
Safety level: No vendor safety framework rating found
- The model card reports cyber capability rows (CyberGym 94.0, ExploitBench 47.9, ExploitGym 17.8, SEC Bench Pro 66.3) but no safety framework rating.
Independent figures: Artificial Analysis, refreshed daily on our AI model benchmarks page.
Xiaomi Model Timeline
1M tokens context
1M tokens context
262k tokens context
262k tokens context
Specifications
Related Tools
Claude Opus 5.5
Claude Opus 5.5, launched 22 September 2026, is Anthropic's current Opus-line flagship, replacing Opus 5. It performs at roughly the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, at $4 input / $20 output per million tokens (both 20% down, cache reads cut 60% to $0.20). It leads on Terminal-Bench 4.0, GDPval-AA v2.1 and Humanity's Last Exam, and is the first Opus model to ship with Fable-5.1-class cybersecurity, biology and distillation safeguards.
Claude Sonnet 5.5
Claude Sonnet 5.5, released 28 September 2026, is the second model in Anthropic's Claude 5.5 family and the faster, lower-cost complement to Claude Opus 5.5. It keeps Sonnet 5's $2 input / $10 output per million tokens ($0.20 cache reads) while generating output over 30% faster, and scores 70.6% on Terminal-Bench 4.0, above Opus 5.5's 66.4%. It has a 1M-token context window, 128K max output, adaptive thinking, and is the first Sonnet with cyber and anti-distillation classifiers.
GPT-6.1 Sol
GPT-6.1 Sol, launched at OpenAI DevDay on 29 September 2026, is OpenAI's mid-tier workhorse model. OpenAI says it delivers nearly the same intelligence as GPT-6 Astra at one-fifth of Astra's token price: $2 input / $10 output per million tokens with cached input cut to $0.10. It scores 75.2% on DeepSWE v1.1 (high effort) and 71.4% on OSWorld 2.0, has a 1.05M-token context window and 128K max output, and is available in the API plus ChatGPT Work and Codex.
