Qwen 3.7 Preview

· ai · Source ↗

TLDR

  • Alibaba previews Qwen3.7-Max-Preview and Qwen3.7-Plus-Preview on Arena, ranking #6 in Text and #5 in Vision labs.

Key Takeaways

  • Qwen3.7-Max-Preview sits #13 overall in Text Arena, with strong vertical ranks: #7 Math, #9 Expert, #9 Software & IT, #10 Coding.
  • Qwen3.7-Plus-Preview ranks #16 overall in Vision Arena; Alibaba is now the #5 vision lab.
  • These are preview releases ahead of the full Qwen3.7 series launch; weights not yet available.
  • Alibaba climbed to #6 lab in Text Arena, signaling continued rapid iteration pace.

Hacker News Comment Review

  • Local runners note Qwen3.6 27B is the first Qwen model reliably runnable on a 3090 with usable context size, but repetition/loop bugs remain a real concern at inference time.
  • Commenters waiting on smaller, Mac-friendly Qwen3.6 variants are now facing a moving target with 3.7 announced before 3.6 small models shipped.

Notable Comments

  • @rspoerri: Qwen3.6 27B is the first model he can run on a 3090 without it losing coherence, though looping is still an issue.

Original | Discuss on HN