W3BStation
Markets
BTC $96,420 +2.34% ETH $3,280 +1.82% SOL $185.40 -0.92% BNB $642.50 +0.45% XRP $2.18 +3.12% DOGE $0.082 -1.50% ADA $1.05 +0.80% AVAX $42.10 +1.15%
BTC $96,420 +2.34% ETH $3,280 +1.82% SOL $185.40 -0.92% BNB $642.50 +0.45% XRP $2.18 +3.12% DOGE $0.082 -1.50% ADA $1.05 +0.80% AVAX $42.10 +1.15%
08/05/2026

Alibaba Releases Qwen3.8-Max AI Model With Open Weights, Trails U.S. Rivals in Benchmarks

What happened: Alibaba unveiled its Qwen3.

Alibaba Releases Qwen3.8-Max AI Model With Open Weights, Trails U.S. Rivals in Benchmarks

What happened: Alibaba unveiled its Qwen3.8-Max model, the first Max-class Qwen to be released with open weights, scheduled to be available on Hugging Face and ModelScope next week. The model features a sparse Mixture-of-Experts architecture with 2.4 trillion total parameters (about 95 billion active per token) and supports multimodal and autonomous coding tasks. According to Alibaba's own benchmarks, Qwen3.8-Max trails Anthropic's Claude "Fable 5" and OpenAI's GPT-5.6 in most text and coding tasks, but leads in document and video multimodal categories.

Why it matters: The open release marks a reversal from Alibaba's previous paid-tier-only stance and could accelerate research in large-scale AI, though the model's size and license terms limit broad self-hosting. While headlines tout "almost matching" U.S. leaders, Alibaba's own scorecard shows it still lags in key benchmarks, especially in coding. The move highlights both the rapid progress and the persistent gap between leading U.S. and Chinese AI labs.

Source: Decrypt