Alibaba’s Qwen3.8-Max launches with 2.4T parameters and strong coding-agent demos
卡尔的AI沃茨 · wechat · 2026-08-04
Qwen3.8-Max arrives with 2.4T parameters, 1M-token context, and a promised open-weight release
The post argues that Alibaba’s Qwen3.8-Max is not just a bigger model, but a faster sparse MoE system with mixed attention that can handle images and long contexts. It also highlights a promised full open-weight release one week after the official launch, plus limited-time pricing cuts on TokenPlan and Qoder.
The real focus is coding-agent performance
The author then reports hands-on tests in Qoder, Alibaba’s terminal-based coding CLI with MCP, Skills, Agents, Hooks, and Memory. The model reportedly:
- built a vertical GalaxyShooter game from an empty directory in about 10 minutes;
- generated a playable browser-based 3D Minecraft-style world with terrain, mining, survival, monsters, inventory, save/load, and respawn;
- modified an existing multi-file AI news product using GSAP + ScrollTrigger without breaking the rest of the app;
- produced a polished Next.js / React / TypeScript launch-site style demo;
- researched 52 macOS tools and turned the findings into an interactive product; and
- assembled a browser-based Windows 11-like desktop with windows, mail, terminal, and file management.
The author’s takeaway is that Qwen3.8-Max already looks strong in end-to-end agentic workflows, especially for long-running coding tasks and multi-file project edits.
Related event: Alibaba releases Qwen3.8-Max, open-sources weights next week(44 posts)→
More from coding & agent
- Agent named 'Ilya' reportedly scores 1.1pt higher on benchmarks — daniel_mac8 · 2026-10-16
- Gemini CLI fix: substring matching inlined unrequested binary files into model context — priyanshu1976 · 2026-10-16
- Building AI agents on real healthcare claims data: the domain work that makes them work — alex_verem · 2026-10-16
- OpenAI stealth-testing mysterious internal model 'Aeon' in Codex desktop app and Dots — cedric_chee · 2026-10-16
- Vibe-coded iPhone Duo sniping bot built with Codex in 17 minutes, 41 tests passing — xiaohu · 2026-10-16
- Long-term multiplayer agent memory tool tops BEAM memory benchmark — CurieuxExplorer · 2026-10-16