Alibaba Launches Qwen3.8-Max, a 2.4T-Parameter Model
Alibaba launched Qwen3.8-Max, a 2.4-trillion-parameter multimodal model benchmarked across 55 agentic, coding, document, spatial, vision and video tasks, with open-weighting for smaller Qwen3.8 variants expected within roughly a week. Alibaba's own figures claim win rates of 96% against Gemini 3.1 Pro and 94% against Opus 4.8, though independent testers note it lands closer to a strong second tier behind Kimi K3 on hands-on evaluation, with some suspicion of benchmark optimization.
Why It Matters
A 2.4T-parameter model landing near the frontier from an Alibaba-blacklisted-for-government-use lab underscores how quickly Chinese open-weight models are closing the capability gap on both benchmarks and real usage.