Alibaba shipped this on September 2 with zero fanfare — no version bump, just a dated snapshot. It debuted at #1 on Code Arena: WebDev with 1691 points: 3 above Claude Opus 5 (Max), 17 above Kimi K3 (Max), 22 above the previous Qwen3.8-Max. First time a Chinese closed-source flagship has beaten Anthropic’s best on this board.
What it is
A 2.4T-parameter MoE model with a 1M-token context window, post-trained specifically for agentic coding: multistep reasoning, tool orchestration, generating full web apps end to end. Same architecture, same weights class as the old Max — the entire 22-point jump came from post-training. It also ranks #1 in the Data & Analytics and Consumer Product categories, not just overall.
The price is the real story
Blended cost runs about $5/MTok ($2 input, $6 output) — a fraction of what Opus 5 charges. It’s live on Alibaba Cloud Model Studio’s API with tool calling and structured outputs, so you can point any coding agent at it today and have it build complete apps. When a post-training refresh alone closes the gap and pricing is this lopsided, “frontier” stops being an American word.
You Might Also Like
- Deepseek tui Tops Github Trending a Claude Code Clone Wired to Deepseeks api
- Code Arena Finally Gives Developers a Fair way to Judge ai Coding Models
- Claude Code Remote Control Just Turned my Phone Into a Coding Terminal and im Weirdly Into it
- Anthropic Just Launched Code Review in Claude Code and 54 of prs now get Real Feedback
- Claude Replay Turns Your Anthropic Claude Code Sessions Into Shareable Video Like Replays
Top AI Product 
Leave a comment