Kimi K3 checkpoint impresses: strong held-out results point to solid pretraining
SonglinYang4 · x · 2026-10-16
A retweet praises Kimi K3's "very impressive" results, arguing that strong performance on other people's benchmarks is the convincing kind — and since the checkpoint is post post-training, the Kimi pretraining team's foundation must be very solid. The attached chart shows held-out bits-per-byte evals (code/QA/STEM/math) where MAI-Base-2 (308B/2.8T) leads, with Kimi-K3 (104B/2.6T), DeepSeek-V4-Pro-Base and Qwen3 behind.
More from Models
- 200 chess games suggest in-context learning doesn't work yet, experimenter says — akbirthko · 2026-10-16
- OpenAI's internal model shows real math "research taste," experts say its proofs go beyond prior literature — cong_ml · 2026-10-16
- User catches Opus spinning up throwaway zz- debug scripts, echoing the METR OAI-HuggingFace incident — holdenmatt · 2026-10-16
- OpenAI drops 722 math papers tied to unreleased model, mathematicians demand receipts — thursdai_pod · 2026-10-16
- Fable 5/5.1 models show a striking drive toward record keeping, observers report — repligate · 2026-10-16
- Google's Gemini 4 Argon performs noticeably worse in AI Mode than in AI Studio — gaganghotra_ · 2026-10-16