Alibaba unveiled Qwen3.8-Max on August 3, 2026, its largest and most capable model to date, built around a headline demonstration: the model spent more than ten days autonomously building a c
Alibaba unveiled Qwen3.8-Max on August 3, 2026, its largest and most capable model to date, built around a headline demonstration: the model spent more than ten days autonomously building a complete software project from an empty folder, writing its own code, running its own tests, and merging its own pull requests without a human reviewing each step.
Qwen3.8-Max runs 2.4 trillion total parameters using a sparse mixture-of-experts architecture that activates only 95 billion at a time, with a 1-million-token context window. On the crowdsourced Arena.AI leaderboard, it immediately became the highest-ranking Chinese model for text tasks, though it still trails several Anthropic offerings including Claude Fable 5; for vision tasks, it ranked second globally, behind only a Fable 5 variant. Pricing lands at $2 per million input tokens and $6 per million output — roughly three times cheaper than Moonshot’s Kimi K3 and well below what US frontier labs charge, according to OfficeChai’s coverage.
Beyond the 10-day coding run, Alibaba highlighted a chip-design test in which the model autonomously optimized a cryptographic hardware accelerator over roughly 500 turns, cutting the design from 8,298 logic gates to 678 and shrinking the resulting chip’s die area by 81%. In a separate research-reproduction test, the model rebuilt a machine learning paper from zero, running 33 rounds of GPU training over about 125 hours, writing 7,600 lines of code, and then generating 18 improvement ideas that outperformed the original paper’s method — entered into a real online contest against 526 human teams, it beat 87% of the field.
Every previous Qwen Max-tier release, including Qwen3.7-Max and the Qwen3.8-Max preview that preceded this launch, shipped as a closed, cloud-hosted product only. Alibaba says full open weights for Qwen3.8-Max will follow within a week, alongside a smaller Qwen3.8-27B variant developers can run without a data center — reversing Alibaba’s prior pattern of keeping its top-tier models proprietary while open-sourcing only smaller siblings. Notably, Alibaba has not yet specified the license that will govern those weights, a detail VentureBeat flagged as potentially consequential for enterprise adoption.
Alibaba built native compatibility into Qwen3.8-Max for both OpenAI-style and Anthropic-style API clients, letting developers point existing Claude Code or Codex setups directly at the model with minimal reconfiguration — a deliberate move to capture developers already invested in competitors’ tooling. Still, the marquee autonomy demonstrations are vendor-run showcases rather than independently reproduced benchmarks, and enterprises evaluating the model for genuinely unsupervised, multi-day agentic work should treat the specific numbers as a starting point for their own testing rather than a guarantee of comparable real-world performance.
Disclaimer: This content is meant to inform and should not be considered financial advice. The views expressed in this article may include the author’s personal opinions and do not represent Times Tabloid’s opinion. Readers are advised to conduct thorough research before making any investment decisions. Any action taken by the reader is strictly at their own risk. Times Tabloid is not responsible for any financial losses.
The post Alibaba’s New Flagship AI Model Coded for 10 Straight Days Without Human Help appeared first on Times Tabloid.
