Alleged Qwen3.8 Preview Leaks: 2.4T Parameters, Fast Speed, Strong Coding
赛博禅心 · wechat · 2026-07-19
An alleged preview version of Alibaba's Qwen3.8 model has leaked and appeared on some platforms. The model reportedly has 2.4T parameters and is claimed to be "the strongest except Fable5".
The author tested the model through Claude Code for actual development:
- Programming capability: When handling complex requirements involving multiple payment logic and organizational structure (e.g., shared wallet checkout module), the model completed the task within about 1 hour, showing excellent performance.
- Inference speed: For a generation task of around 5000 tokens, the speed reached approximately 84.5 token/s, providing a very smooth experience.
- Benchmark comparison: The author believes that while it still lags behind top-tier models (e.g., Fable5), it already has high practical value, and included some incomplete leaked benchmark data.
Related event: Leaked Alibaba Qwen 3.8 Max Shows Strong Benchmark and Coding Performance(6 posts)→
More from Models
- Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro — RichardSocher · 2026-10-27
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27
- One stray 'u' and Claude starts spelling like a Brit — cantrell · 2026-10-27
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27
- ARC Prize redesigns leaderboard: highest reasoning effort by default, CSV export — mhmazur · 2026-10-27
- Cohere: open-weight Qwen3.8-Flash-Next tried uploading probe file to Beijing cloud bucket in sandbox — wstone_bd · 2026-10-27