Alibaba Launches 2.4T Parameter Qwen3.8-Max, Codes Autonomously for 16 Days
智东西 · wechat · 2026-08-03
Alibaba has officially released its flagship model, Qwen3.8-Max, featuring 2.4 trillion total parameters (95B active). It is the first Max-tier model in the Qwen series planned for open-weight release. Alibaba's Hong Kong shares surged 7.26% on the announcement.
The new model heavily focuses on end-to-end completion of complex, long-horizon tasks. In extreme tests demonstrated by the Qwen team, Qwen3.8-Max showcased astonishing autonomous working capabilities:
- 16 Days of Autonomous Coding: Managed a GitHub project state without human intervention, accumulating 265 commits, 127 PRs, and 151 Issues.
- Paper Reproduction & Improvement: Spent 125 hours reproducing a research paper from scratch, writing 7,600 lines of code, and autonomously proposing 18 improvement strategies, ultimately boosting accuracy on AIME 2024 by an additional 2.7 points.
- Algorithmic Competitions: Defeated 458 human teams (out of 526) in a real-world multimodal dialogue intent recognition challenge within 24 hours.
- Chip Design: Optimized a GCD/RSA hardware accelerator over 500 interaction rounds, reducing logic gates from 8,298 to 678, shrinking the overall area by 81%.
In authoritative benchmarks, Qwen3.8-Max surpasses or closely trails the latest flagship models like GPT-5.6 Sol and Claude Fable 5 on OSWorld-Verified (computer operation) and PaperBench. However, it still lags slightly behind in SWE-bench Pro and TerminalBench 2.1.
API pricing is set at 12 RMB/MTok input and 36 RMB/MTok output domestically, and $2/MTok input and $6/MTok output internationally. Model weights will be available for download on HuggingFace and ModelScope next week.
Related event: Alibaba releases Qwen3.8-Max, open-sources weights next week(44 posts)→
More from coding & agent
- DeepMind turns coding agent Antigravity into a science tool, surfacing novel antimicrobial peptide in one session — pushmeet · 2026-10-16
- Cross-review coding workflow: Claude writes the plan, Codex audits it — theteknosaur · 2026-10-16
- Simon Willison: code mode is the future of tool calling, if it can be sandboxed — springrod · 2026-10-16
- Normalizing Claude's schema-free MCP output before it hits downstream n8n nodes — snowingbol · 2026-10-16
- Agent named 'Ilya' reportedly scores 1.1pt higher on benchmarks — daniel_mac8 · 2026-10-16
- A plain notes.md fixed my agent's cross-session amnesia better than chat history — rahul_watertech · 2026-10-16