Research agent Épi sets 120 records, narrows n-queens constant interval 84x
dair_ai · x · 2026-10-21
Épi is a self-improving research agent that sustains investigations across model sessions, combining persistent research history, hierarchical memory, versioned branches, and independent review—with a fixed verifier scoring every output so the agent can't alter its own judgment.
It set 120 record-setting results: 20 numerical bound records across 12 families and 100 construction records across 8 families, including an 84x narrower interval for the n-queens constant, a >360-million-fold tighter rigorous interval for the quartic Feigenbaum attractor, exact dual certificates proving the Cohn-Elkies bound can't be sharp in dimensions 10 and 11, and a computer-assisted proof of a Tuza's conjecture case for split graphs. A lightweight variant, Épi-turbo, writes solvers for routing and scheduling: on 30 FrontierOR tasks it scores 5.4% higher than Codex at 64% of the model cost, using the same model and reasoning effort.
More from coding & agent
- Burkov: without explicit checklists, coding agents hand you 80%-done work — burkov · 2026-10-21
- Laude Institute launches Slingshots FOUR: 14 projects across 8 top institutions — a1zhang · 2026-10-21
- Compaction hooks in practice: keep last n messages verbatim, summarize the rest for full agent continuity — MoonL88537 · 2026-10-21
- The blank prompt box hides real operational burden: context, workflow and supervision fall on users — _AustinCalvert_ · 2026-10-21
- DeepMind paper: self-evolving agents forget as they learn — fastest learner GEPA also breaks old tasks most — rohanpaul_ai · 2026-10-21
- arXiv: Are Self-Evolving Agents Continual Learners? Plasticity comes at the cost of stability — rohanpaul_ai · 2026-10-21