DeepSeek-V4.1-Flash paper hints at continuously tunable reasoning effort from 1 to 100
zainhas · x · 2026-09-10
- A post circulating on X points to DeepSeek's paper describing DeepSeek-V4.1-Flash supporting a continuously controllable reasoning effort from 1 to 100, instead of the usual low/medium/high presets.
- The author notes he has never seen this design in any model and thinks it "complicates things."
- Details live in the linked paper; the claim is not yet officially confirmed, so treat it as a credible leak.
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