← Blog

Qwen3.8-Max takes fourth place on coding leaderboard – but the weights aren't out yet

Mads Kristiansen

CTO, Liviate

China's open-weight wave continues

Alibaba has just pulled back the curtain on Qwen3.8-Max, and the numbers speak for themselves: fourth place on the Frontend Code Arena with 1,668 points – beaten only by Anthropic's Claude Opus 5 and Moonshot AI's Kimi K3.

It's the first time an open-weight model from Alibaba has cracked the top 5 on this leaderboard, cementing a trend we've watched all year: Chinese open-weight models are closing in on frontier-level performance at a rapid pace.

Two models, one release

Alibaba hasn't just thrown a single model over the wall. Alongside Qwen3.8-Max comes a smaller 27B dense model – both announced as open weights. That gives developers a choice between raw power and practical deployability depending on their needs.

The market reacted promptly: Alibaba's stock rose 6% following the announcement.

An important nuance: announced ≠ released

Here we need to be precise – something that matters more and more in the open-weight landscape:

At the time of writing, the Qwen3.8-Max variant had NOT yet been released as official weights from Alibaba's own Hugging Face organization. The API was live ($2/$6 per million tokens), but the actual weight files were still missing – only third-party distills and unofficial quantized copies of the smaller 27B model had surfaced.

This illustrates exactly the debate around [no weights]: a model can be announced as open weight without having actually delivered its weights yet – much like calling something "open binary" when all that's released is a compiled binary, with no source code or the steps behind it.

Why this matters for European teams

When models like Qwen3.8-Max actually ship as open weights, they can be hosted anywhere – including European data centers under GDPR. You're not locked into a US or Chinese API.

At Liviate, we're seeing rising demand from teams who want to run their own open-weight models on European infrastructure – both the big flagship models and the smaller production-ready ones – while we help them navigate the gap between what's announced as open and what can actually be downloaded and deployed responsibly.

Want to talk about what's actually blocking your AI initiatives?

Mads Kristiansen is happy to have a no-obligation chat.

Book a meeting →