[AINews] Qwen 3.8 Max(2.4T) and 27B, new open weights models for Coding and Cowork

Alibaba’s Qwen team announced Qwen3.8-Max, a 2.4-trillion-parameter model focused on coding, long-horizon agents, multimodal work and “cowork,” and said its open weights will be released next week alongside Qwen3.8-27B. The announcement removes recent uncertainty about whether Qwen would continue producing major open-weight models, although the weights had not yet been released in the reporting. Alibaba is offering Qwen3.8-Max through its API at $2 per million input tokens, $6 per million output tokens and $0.25 per million cached tokens. The company claims capabilities including more than 10 days of unattended coding, autonomous research runs, chip-design optimization, extended e-commerce simulations and native visual feedback in agent workflows. Third-party reporting lists a 1-million-token context window, up to 128,000 output tokens and about 95 billion active parameters per token, though some of those specifications come from secondary sources. Early external evaluations were strong: Arena placed the model fourth in Frontend Code Arena and second in Vision Arena, while Vals reported a 66.1 score, 87.3% on SWE-bench and 67.4 on Terminal-Bench 2.1. Vals also reported that it tr

Follow Latent Space to make it a durable For You signal.