Épi research agent sets 120 records, beats Codex by 5.4% on FrontierOR at 64% of the cost
Arindam_1729 · x · 2026-11-02
The paper "Épi: Towards a Persistent, Self-Improving Agent for Autonomous Research" tackles the core limitation that most agents forget everything when a session ends. Épi maintains persistent research history, hierarchical memory, versioned branches, and independent review, letting experiments span session boundaries with a fixed verifier scoring outputs.
- 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 nn-queens constant and a >3.6-million-fold narrower rigorous interval for the quartic Feigenbaum attractor dimension
- Exact dual certificates prove the Cohn–Elkies LP bound cannot be sharp in dimensions 10 and 11
- A computer-assisted proof of Tuza's conjecture for split graphs with an 8-vertex clique part and ≤3 triangle-active neighborhood types
- Épi-turbo extends the approach to algorithmic coding: on 30 FrontierOR tasks it beats Codex by 5.4% total score at 64% of the cost, same model
Related event: Épi agent sets 120 records with persistent memory(2 posts)→
More from coding & agent
- RL weight-update pause time doesn't depend on datacenter distance, authors say — gharik · 2026-11-03
- Hermes adds one-command migration for Grok Bot and OpenBot users — NousResearch · 2026-11-03
- A Weekend of Nudging Got Claude to Write Workable Comedy, Plus a Better-Than-Random Discriminator — ivan_bezdomny · 2026-11-03
- Developer spends a weekend teaching Claude comedy writing, builds a joke discriminator — ivan_bezdomny · 2026-11-03
- Scenario open-sources 3D OS: 80 agent skills that take a sketch to game-ready 3D assets — Scobleizer · 2026-11-03
- Solo dev ships 4-player co-op game about 1287 Mongol invasion, built heavily with Claude — Ok-Pirate5732 · 2026-11-03