Princeton's QUEEN: a 4B model that plays grandmaster-level chess and explains its moves in plain language
jiqizhixin · x · 2026-10-13
Princeton's Danqi Chen team presents QUEEN, a 4B-parameter model that plays chess near grandmaster level while narrating its reasoning in natural language, bridging engines that play but can't explain and LLMs that explain but can't play.
- Architecture: an Lc0 chess network encoder feeds layer embeddings through Flamingo-style gated cross-attention blocks injected into an instruction-tuned SmolLM3-3B decoder, with all original parameters frozen during domain adaptation.
- Cost matters: the paper notes a full game with GPT-5.6-Sol's high-reasoning tier can cost over $15. With only 1,899 grandmasters worldwide (as of Oct 2), the task remains extremely rare and hard.
More from Research
- AgentSWE benchmark: no coding agent comes close to delivering the agents it's asked to build — Jiahao Wang · 2026-10-13
- Schema agent hits ~99% RHAE on ARC-AGI-3 public set, full paper and code released — wzenus · 2026-10-13
- World model research: more data won't suffice—squeeze more from it; even intuitive physics is still out of reach — mangahomanga · 2026-10-13
- Indie dev builds own serving library to train SVG model, finds evolution loses to SGD — cephaloform · 2026-10-13
- Harvard & NTT Research: cooperating AI agent populations hit an Allee-effect takeoff threshold — Hidenori8Tanaka · 2026-10-13
- RT-Super trains tumor segmentation from radiology reports, lifting Dice from 16 to 30 without masks — maier_ak · 2026-10-13