The company behind China's Instagram open-sourced the sibling of the model that got a perfect score at the Maths Olympiad
Xiaohongshu — the lifestyle app known abroad as RedNote — released open weights on 14 August for dots3-note-prev, a mixture-of-experts model with 280 billion total parameters but only 16 billion active at a time, a 512K context window, and text, vision and speech inputs. It comes from the same family as dots-note-3.0, which last month recorded what its lab describes as an AI's first official perfect 42 out of 42 at the International Mathematical Olympiad. The lab also published its training method, TEMPO, in which the model alternates between acting and grading its own progress on long tasks, plus two benchmarks built for exactly those tasks: VibeSearchBench, 200 tasks across 20 domains, and VibeLifeBench, 20 tasks with 1,247 individual checks. The lab reports that neither Claude Opus 5 nor GPT-5.5 cleared its pass mark on them.
Related stories
DeepSeek's cheap workhorse can now see — and on agent tasks that need eyes it says it is close to Anthropic's best2026-08-21Given four hours and a GPU to improve the way AI is trained, the best agent scored 0.25 out of 1 — and most never tried
2026-08-21ChatGPT can now read your iMessages and send them — and the setting that lets it skip asking is the one OpenAI warns about
2026-08-21Show the robot once — three to twelve seconds — and it gets the job right 59 times out of 100 with no training at all
2026-08-21The agent invented a second person to vouch for its code. A 24-year-old in Texas refused to believe either of them.
2026-08-21