Did OpenAI distill DeepSeek's gains? Benchmark shows OpenAI's Luna scoring just 38
teortaxesTex · x · 2026-10-18
teortaxesTex pushes back on the claim that DeepSeek's V4.1 → V4.1-1017 improvement came purely from distilling OpenAI models, posting a benchmark chart where OpenAI's Luna scores only 38 — below several DeepSeek models — despite running on superior hardware yet being slower. His jab: if OpenAI can't distill a Luna that beats V4.1, how small must Luna be, and how fat are OpenAI's margins?
More from Fun
- Borges's nine stories turned into walkable 3D worlds with AI model Fable 5.1 — emollick · 2026-10-18
- Soham Parekh, the engineer who juggled multiple AI startup jobs, made it into a movie — kingslyj · 2026-10-18
- Luo Yonghao meme blind-box figures spotted in stores at 79 yuan each — yihui_indie · 2026-10-18
- Xiaohu: AI comment bots are even more disgusting than spam porn bots on X — xiaohu · 2026-10-18
- 13-year-old nails Bay Area parenting: kids must be both Steph Curry and Elon Musk — salgar · 2026-10-18
- AI-generated stage projections keep repeating "For a Brighter Tomorrow" — why? — Awkward-Concept5736 · 2026-10-18