Breaking

Xiaomi MiMo-V2.6-Pro Enters Code Arena's WebDev Top Tier With MIT-Licensed Open Weights

September 21, 2026 at 17:08 EDT

Foundation Models
AI Agents
Software Dev & Coding

Xiaomi's MiMo-V2.6-Pro has appeared near the top 10 of Code Arena's WebDev leaderboard with a 1,628 AutoEval score, putting the MIT-licensed open-weight model in the same scoring band as leading proprietary coding systems. The result lands as Xiaomi publishes and open-sources the broader MiMo-V2.6 series, including MiMo-V2.6-Pro and MiMo-V2.6-Flash, with omnimodal inputs, long-context agent support and API availability.

September 2026 · Xiaomi MiMo

MiMo-V2.6-Pro enters WebDev's top tier

The MIT-licensed open-weight model reached a 1,628 AutoEval score on Code Arena WebDev, tying Claude Fable 5 High and edging Hy4-preview.

1,628

Code Arena WebDev AutoEval

+153

Points over MiMo-V2.5-Pro

Token context window

MIT

Open-weight license

WebDev score jump above 1,400

1,475

MiMo-V2.5-Pro

1,628

MiMo-V2.6-Pro

Columns show leaderboard points above 1,400, making the 153-point gain visible while keeping the scale compact.

Why developers noticed

Open weights, an MIT license and a high WebDev score make Pro a candidate for harder planning, review and agentic coding workflows.

What to watch

The leaderboard row is still AutoEval, several Xiaomi benchmarks are in-house, and the 1.02T-parameter MoE is demanding to serve locally.

Release path

Public RL run

→

Open weights

→

API and tools

→

WebDev top tier

Continue reading

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