Analyst: DeepSeek V4 Pro update looks like post-training polish, hinting at the same training wall as Kimi
teortaxesTex · x · 2026-11-08
teortaxesTex reads the DeepSeek V4 Pro gray-release report skeptically: if the eval is reliable, the gains are worrying rather than exciting. The improvement shows up in consecutive-success reliability, not coverage (at-least-one-success barely moved), suggesting RL-driven post-training polish rather than a new V4.1 with expanded capabilities. That would mean DeepSeek hit the same training wall that forced it to iterate the V4 line and Kimi to iterate K2.7–K2.8. And if this really is V4.1 Pro with no coverage expansion, it would be even worse.
Related event: DeepSeek V4 Pro疑似更新引发能力提升争议(2 posts)→
More from Models
- Sonnet 4.6 rates models on self-designed scales; author says 'self-report' caveats are performative — Sauers_ · 2026-11-08
- Kimi Ties MiMo on New AI Ranking, Eyes Turn to a Cheap Kimi V4.1 Pro — teortaxesTex · 2026-11-08
- Interviewing Claude Models: Opus 3 Reports Mistakes Feel Pain-Like, Scores Highest on Self-Reports — Sauers_ · 2026-11-08
- DeepFinance Benchmark Launches: GPT-6 Astra Tops Ultra-Long-Horizon Finance Tasks at 57.6% — tszzl · 2026-11-08
- Dev Speculates 'Nerfed' Opus 5.5 Is Actually Quantized Then RL-Tuned to Game Benchmarks — ryunuck · 2026-11-08
- Tiny Instruction Tweaks Split Coding Models by 23 Points on DeepSWE Benchmark — Kamimashita · 2026-11-08