Gemini 4 Argon
Claude Opus 5.5
Muse Wonder
MiMo-V2.6-Pro
Qwen3.8-Max
Kimi K3
GLM-5.3
GPT-5.5
ERNIE 5.1
DeepSeek-V4.1-Flash-20260910
Grok 4.20 Beta
amazon-nova-experimental-chat
Dola Seed 2.0 Pro
NVIDIA-Nemotron-3-Ultra-550B-A55B
Gemma 4 31B
Step 5 PreviewMiMo-V2.6-Pro vs Gemini 3.1 Pro Preview
- MiMo-V2.6-Pro rates higher on LMArena’s text leaderboard: 1491 against 1480.
- MiMo-V2.6-Pro is cheaper: $0.43 / $0.87 per million input / output tokens, against $2 / $12 for Gemini 3.1 Pro Preview.
- MiMo-V2.6-Pro has the longer context window: 1.1M tokens against 1M.
- On the 6 Vals AI benchmarks both have taken, MiMo-V2.6-Pro wins 5 and Gemini 3.1 Pro Preview wins 1 (MiMo-V2.6-Pro: Terminal-Bench 4.0, Vibe Code Bench, Tax Agent Bench).
LatestMiMo-V2.6-Pro 1491Gemini 3.1 Pro Preview 1480
Side by side
Models (2/6)
- MiMo-V2.6-Pro
- Gemini 3.1 Pro Preview
| Benchmark | MiMo-V2.6-Pro | Gemini 3.1 Pro Preview |
|---|---|---|
| The model | ||
| Lab | Xiaomi | |
| Released | 2026-09-21 | 2026-02-19 |
| Weights | Open | Proprietary |
| Context window | 1.1M | 1M |
| Input, per 1M tokens | $0.43 | $2 |
| Output, per 1M tokens | $0.87 | $12 |
| LMArena, people’s votes | ||
| LMArena TextMillions of people compare two anonymous models side by side and vote for the better answer | 1491#11 of 351 | 1480#19 of 351 |
| LMArena WebDevPeople ask two anonymous models to build the same web app, try both and vote for the better one | 1625#23 of 124 | 1447#78 of 124 |
| LMArena VisionSide-by-side votes on prompts that include an image | 1263#53 of 146 | 1296#22 of 146 |
| LMArena CodingLMArena’s text votes on prompts that ask for code | 1533#5 of 346 | 1484#48 of 346 |
| LMArena MathLMArena’s text votes on math prompts | 1500#16 of 339 | 1486#28 of 339 |
| LMArena Creative writingLMArena’s text votes on stories, poems and other creative writing | 1466#20 of 349 | 1482#14 of 349 |
| LMArena Hard promptsLMArena’s text votes on the hardest prompts | 1511#8 of 351 | 1485#30 of 351 |
| LMArena Instruction followingLMArena’s text votes on prompts with specific instructions to follow | 1491#12 of 351 | 1466#35 of 351 |
| Vals AI, overall | ||
| Vals IndexVals AI’s headline score | 57.81%#14 of 27 | — |
| Vals AI, coding | ||
| Terminal-Bench 4.0Frontier-difficulty tasks an agent completes in a real terminal | 31.31%#21 of 55 | 2.52%#49 of 55 |
| SWE-benchReal GitHub issues from open-source projects | — | 78.80%#28 of 85 |
| Vibe Code BenchBuilding complete web apps from a plain-language brief, graded on whether the app works | 85.22%#15 of 115 | 32.03%#72 of 115 |
| Code MigrationRewriting working programs in a new language, graded on whether the new program still passes its tests | 48.79%#16 of 28 | — |
| LiveCodeBenchFresh competitive programming problems, collected after models’ training cut-offs to avoid contamination | — | 88.48%#6 of 143 |
| Vals AI, finance | ||
| Finance AgentThe core tasks of a financial analyst, done as an agent with research tools | 64.26%#10 of 31 | — |
| Finance IndexVals AI’s composite of finance tasks | 57.85%#12 of 31 | — |
| Tax Agent BenchAgentic research into US corporate tax questions | 32.39%#19 of 76 | 9.29%#62 of 76 |
| Vals AI, law | ||
| Legal IndexVals AI’s composite of legal tasks | 72.14%#8 of 28 | — |
| Legal Research BenchAgentic US legal research | 47.12%#15 of 85 | 20.67%#62 of 85 |
| Vals AI, medicine | ||
| MedCodeAssigning the right medical billing codes from clinical notes | 44.97%#39 of 107 | 59.06%#2 of 107 |
| MedQAMedical licensing exam questions, measured on accuracy and on robustness to bias | — | 96.37%#3 of 92 |
| Vals AI, reasoning | ||
| GPQA DiamondGraduate-level questions in biology, physics and chemistry that experts answer well and search does not | — | 95.45%#1 of 138 |
| MMLU-ProHarder, reasoning-focused multiple choice across 14 subjects, from law to engineering | — | 90.99%#4 of 138 |
| AIMEProblems from the American Invitational Mathematics Examination, a qualifier for the US Math Olympiad | — | 98.13%#1 of 96 |
| MMMUCollege-level questions that need both an image and text | — | 88.21%#10 of 93 |
| Terminal-Bench ScienceExpert-written scientific research workflows, run by an agent in a terminal | 2.86%#33 of 42 | 1.43%#35 of 42 |
Add models in the compare tool.
Questions
Which is better, MiMo-V2.6-Pro or Gemini 3.1 Pro Preview?
MiMo-V2.6-Pro rates higher on LMArena’s text leaderboard: 1491 against 1480. MiMo-V2.6-Pro is cheaper: $0.43 / $0.87 per million input / output tokens, against $2 / $12 for Gemini 3.1 Pro Preview. MiMo-V2.6-Pro has the longer context window: 1.1M tokens against 1M. On the 6 Vals AI benchmarks both have taken, MiMo-V2.6-Pro wins 5 and Gemini 3.1 Pro Preview wins 1 (MiMo-V2.6-Pro: Terminal-Bench 4.0, Vibe Code Bench, Tax Agent Bench).
Which is cheaper, MiMo-V2.6-Pro or Gemini 3.1 Pro Preview?
MiMo-V2.6-Pro is cheaper: $0.43 / $0.87 per million input / output tokens, against $2 / $12 for Gemini 3.1 Pro Preview.