• Sources: Qwen (@Alibaba_Qwen), HN discussion
  • Summary: The Qwen team announced Qwen 3.8, a 2.4T-parameter model it describes as "launching and going open-weight soon", with a Qwen3.8-Max-Preview available for early testing on its Token Plan, Qoder, and QoderWork surfaces. The post positions the model as "second only to Fable 5" among frontier models. It gives no benchmark numbers or method, no license, no weight files, and no release date at announcement.
  • Comments: HN commenters read the announcement as a direct response to Moonshot's 2.8T-parameter Kimi K3, whose open weights are due 2026-07-27, and note the shift from smaller value-tier models toward very large, slow, high-parameter open releases. Several flag that the "second only to Fable 5" claim arrives with no benchmarks to check.
  • Why it matters: Two Chinese labs racing to publish multi-trillion-parameter open weights within the same window sets the near-term ceiling for self-hostable model scale, even before any of the claims are independently measured.
  • Follow-up: Watch for the Qwen 3.8 weights, license, a technical report, and independent benchmarks.

send feedback on this story