Arena Code Arena: MiMo-V2.6-Pro WebDev AutoEval Record

One-sentence conclusion

Arena's Code Arena | WebDev overall leaderboard includes mimo-v2.6-pro with an AutoEval score of 1628 (+18/-18), but it does not publish a vote count or rank, so this only shows that it was included in the WebDev automated evaluation leaderboard; 1628 must not be treated as a blind-test ranking.

Suitable scenarios

• Suitable tasks: Front-end web development, as well as agentic coding workflows that require multi-step reasoning and tool use.

• Unsuitable tasks: This record cannot establish the model's quality for general text, vision, multimodal tasks, long-form writing, or real-world project delivery; nor can it establish human preference.

• Applicable model version: mimo-v2.6-pro; the page's model alias is mimo-v2.6-pro.

• Applicable client, agent, or API: Arena Code Arena's WebDev evaluation environment; the page does not publish the specific harness, tool schema, system prompt, or API provider configuration.

• Recommended reasoning level and parameters: The page does not publish the reasoning level, temperature, sampling parameters, or tool configuration; these should not be filled in from assumption.

Test method

1. Open Arena's official Code leaderboard: https://arena.ai/leaderboard/code.

2. Use the page's default Ranking, Overall, and Models views, and record the leaderboard title, page date, total vote/session statistics, and total number of models.

3. Locate the exact alias mimo-v2.6-pro in the model table. Do not merge mimo-v2.5-pro, mimo-v2-pro, or mimo-v2-flash into this record.

4. Record Rank, Rank Spread, Score, Votes, price, and context separately. If a field shows N/A, retain N/A; do not substitute a neighboring model's data or official promotional data.

Original evidence and data

Leaderboard environment

mimo-v2.6-pro row

The key distinction is that the leaderboard header shows 679,295 votes, while mimo-v2.6-pro itself has N/A for Votes, and its Rank is N/A with an AutoEval marker next to the score. Therefore, it cannot be written as “MiMo ranked Xth in Arena's blind test,” nor can 1628 be used for a strict win/loss comparison with models that have publicly reported Votes.

Scope and limitations

• This is an independent evaluation record on Arena's official page, but the page does not specify the AutoEval task set, sample count, evaluation script, tool version, reasoning level, or provider routing. These omissions limit reproducibility and horizontal comparison.

• The leaderboard only covers the Overall view of Code/WebDev; this record cannot be generalized to general chat, visual understanding, or full multimodal capabilities.

• Price and context length are model metadata displayed by the leaderboard, not quality results from this AutoEval, and may change over time.

• On the same page, other models' publicly reported vote-based scores and MiMo's AutoEval score are different types of evidence; they should be presented in separate columns in a report.

• During collection, Arena's official X account @arena and the latest from:arena MiMo search entry point were checked. The X timeline remained in a loading state and showed ChunkLoadError/fetchError; no verifiable post body was obtained. Therefore, no X search snippet or hearsay was included in the conclusions.

• Arena's Text page did not show the exact alias mimo-v2.6-pro during collection, while the Agent page showed only Mimo V2.5 Pro. Neither can serve as a Text/Agent score for MiMo-V2.6-Pro, and neither should be merged with this WebDev AutoEval record.

Source excerpts or observations (brief excerpts for compliance only)

• Leaderboard header notice: mimo-v2.6-pro's AutoEval score is now on Arena.

• Original model-row fields: mimo-v2.6-pro · Xiaomi · MIT · 1628 · +18/-18 · AutoEval · N/A · $0.43/$0.87 · 1M.

• The page description explicitly defines this leaderboard as covering front-end web development tasks, including agentic coding workflows that require multi-step reasoning and tool use.