Table of Contents
Alibaba just pulled off one of the weirdest flexes I’ve seen from a Chinese AI lab: it caught up to Anthropic’s Claude Fable 5 without even bothering to bump the version number. Qwen-3.8-Max-0902 is still just an “0902” update, and yet here it is, matching Fable 5’s capabilities roughly two months after that model launched.
Qwen-3.8-Max-0902 tops Code Arena WebDev
The model has 2.4 trillion parameters and comes in cheap: $2 per million input tokens, $6 per million output tokens, plus small change for cache hits. That pricing helped it land the number 1 spot on Code Arena WebDev, and it’s placing well across the board too, first in Data & Analytics and Consumer Product, second in Brand & Marketing, Gaming, and Simulations, and third in Content Creation Tools and Reference-Based Design.
Qwen-3.8-Max-0902 is the best deal out there right now, hands down. It scores 1,629 on Code Arena WebDev and only costs $2 per million input tokens. Claude Opus 5 (Max) scores a bit higher at 1,688, but costs $20 per million tokens, ten times more. That’s a tiny score bump for a massive price jump.
This also comes barely two weeks after Alibaba’s Qwen-3.8 27B, an open-weight model with coding chops close to Opus 4.5 that somehow runs on a single MacBook. Alibaba is clearly not slowing down.
Naturally, catching up to Fable 5 this fast is going to raise distillation eyebrows. I won’t pretend to know if that’s what happened here, especially since Fable 5 launched with guardrails specifically built to stop that kind of thing. It’s also a bit rich given Alibaba reportedly banned its own employees from using Claude Code, branding it a “high-risk” tool.
Meanwhile, Alibaba’s Ulanqab data centre in Mongolia is quietly becoming the backbone of all this. Near-constant sunshine keeps power costs down to roughly 0.32 to 0.35 yuan per kilowatt-hour; that’s about $0.05 per kWh. Cheap compute is clearly doing a lot of the heavy lifting here.
Version number or not, this is a serious leap.
For more updates like this, check out the gaming news section!
More Stories
Anthropic beats OpenAI with the cheaper Claude Fable 5.1 model
OpenAI Gives Free ChatGPT Users Unlimited Text Chats With GPT-5.6 Luna
Kimi K3’s reasoning is suspiciously close to Claude’s