Breaking

Xiaomi's MiMo-V2.6-Pro Climbs to No. 8 on Design Arena After Open-Weight Release

September 21, 2026 at 17:42 EDT

Foundation Models
AI Agents
Open Source

Xiaomi's MiMo-V2.6-Pro has jumped into the top tier of AI design leaderboards, reaching an Elo rating of 1338 and No. 8 overall on Design Arena, including No. 3 among open-weight models. The result marks a 54-point Elo gain and a 22-place rise from MiMo-V2.5-Pro, with the model also reaching No. 4 overall in Website and No. 2 among open-weight models in that arena.

Sep. 21, 2026 · Xiaomi MiMo

MiMo-V2.6-Pro breaks into the design top 10

The open-weight model reached 1338 Elo and No. 8 overall on Design Arena after Xiaomi's scaled reinforcement-learning release.

1338

Design Arena Elo

Overall rank

Open-weight rank

Context tokens

DeepSWE v1.1 improved during the RL run

58.4Before Pro RL

72.6After Pro RL

The Pro run gained about 14 points on an out-of-sample software-engineering benchmark while the release also reported 30 RL steps and roughly 750,000 total trajectories across Pro and Flash.

Why developers care

Open weights, multimodal inputs and strong design-agent scores make the model attractive for coding agents, web generation and long-context automation.

What still needs proof

Crowdsourced Elo is not a correctness test, and a 1.02T-parameter MoE remains demanding for teams that want to self-host or verify outputs at scale.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.