Qwen3.8-Max Announced: 2.4 Trillion Parameters for Days of Autonomous Work
Published August 3, 2026
The Qwen team has unveiled Qwen3.8-Max, its most powerful AI model yet — built not just for better answers, but to work autonomously on complex projects for hours or even days.
The model features 2.4 trillion parameters with around 95 billion active per step. Particularly noteworthy: for the first time, Qwen plans to release the weights of a Max-class model openly on Hugging Face and ModelScope.
Codes autonomously for 16 days
In the demo project "oh-my-cli", Qwen3.8-Max worked fully autonomously for about 16 days: new requirements were turned into GitHub issues, implemented by agents and verified through tests and CI checks. The result: 265 commits, 127 pull requests and 151 issues — with the model continuously improving its own development process.
Reproduces research — then improves it
Given only a research paper and GPU access, the model rebuilt the entire pipeline from data processing to evaluation: around 7,600 lines of code and 33 training runs in 125 hours. It then tested 18 of its own improvement ideas, ultimately beating the paper's original method by 2.7 percentage points on the AIME24 math benchmark.
Better than 87 percent of human teams
In the WWW2025 Multimodal Dialogue Intent Recognition Challenge, Qwen3.8-Max built a multimodal solution within 24 hours and raised its accuracy from 0.60 to 0.853 across 45 submissions — outperforming 458 of 526 human teams.
Availability
Qwen3.8-Max is available via QwenCloud, with APIs compatible with OpenAI's Chat Completions and Responses APIs as well as Anthropic's protocol. A reasoning_effort parameter (xhigh, medium, low) lets developers trade accuracy for speed and cost. The open weights are expected to follow on Hugging Face and ModelScope.
Assessment
The showcased experiments are technically impressive — but many benchmarks are vendor-internal and not independently verified. That is exactly why the announced open-weight release matters: it would let researchers and companies put Qwen's claims to the test.
Source: Qwen Team – Qwen3.8-Max: A New Bar for Coding and Cowork
