Peking University

Research on the record

• 2026-10-24ByteDance Seed and Peking University build exact total spin into neural network wavefunctions, and resolve the spin ladders of iron-sulfur clustersLevel 3 · Checked against existing evidence
• 2026-10-23Refining the reconstruction with a neural network recovers atom-by-atom electron tomography at a sixth of the electron dose, on one re-read tilt seriesLevel 2 · Computed
• 2026-10-08A world model learns to skip ahead in atomic simulations by predicting which atoms move and how much time passed, with the physics bolted on as hard constraintsLevel 2 · Computed
• 2026-10-06A Peking University team and coding agents formalize the Poincaré conjecture in Lean: about 2.8 million lines, a little over two weeks for the main phase, about $25,000Level 2 · Computed
• 2026-09-29CARAT tests whether a materials LLM reasons over a crystal structure or copies the answer printed in its inputLevel 2 · Computed
• 2026-09-15A model trained on the electrocatalysis literature ranks seven rare-earth dopants, and the lab confirms its top fourLevel 4 · Tested in the physical world
• 2026-09-14Peking University's OpenAI4S, an open-source research agent that keeps every step as rerunnable code, scores 7.83 against 5.7 to 6.4 for Claude Code on 36 science tasksLevel 2 · Computed
• 2026-09-09Westlake University and DP Technology publish ProteinTalks in Nature, a virtual cell trained on drug-treated breast cancer proteomes whose drug picks were tested on three patients' organoidsLevel 4 · Tested in the physical world
• 2026-03-01RealPXRD-Solver, led by DP Technology, reads crystal structures off measured powder diffraction patterns, and solves 39 database entries whose structures had never been reportedLevel 3 · Checked against existing evidence
• 2025-10-23METiS says MTS-004, a tablet whose formulation its own AI tool designed, met the primary endpoint of a 264-patient Phase 3 in uncontrollable laughing and cryingLevel 5 · Used in practice

Graded by how far the evidence went, not how good the result is. See the methodology.

Read on AI4S.fyi

• 2026-09-26MatSemNet: a literature-trained model ranks seven rare-earth dopants, and the lab measures the same top fourBrief

People

Only people the AI4S.fyi record has named: in a round, a timeline entry, commentary, an interview or a sourced role. Not the full team.