Alibaba unveils Qwen3.8-Max, 2.4 trillion parameters, open weights
The model ranks just behind Claude Fable 5 on a reference benchmark, ahead of Claude Opus 4.8. If the open weights actually ship, you could run something close to the market leader without depending on a proprietary API.
practitioners › 81.25 on KingBench 3, preview available on Token Plan and Qoder, open weights announced but not yet released.
Qwen (annonce X) →