Épi: a persistent research agent reports 120 record-setting results

What the research found

Épi is an autonomous research agent that persists investigations across sessions and reports 120 record-setting math and algorithm results, plus an Épi-turbo coding variant that beats Codex by 5.4% on 30 FrontierOR tasks at 64% of the cost; this is an arXiv preprint with no code release confirmed in the text.

Claims and how well the evidence supports them

1. Épi produced 120 record-setting results: 20 numerical bound records across 12 families and 100 construction records across eight families 99%
2. On 30 FrontierOR tasks, Épi-turbo achieves 5.4% higher total benchmark score than Codex at 64% of its model cost, using the same underlying model 98%
3. Épi gives a computer-assisted proof of Tuza's conjecture for split graphs with an eight-vertex clique part and at most three triangle-active neighborhood types 98%
4. Exact dual certificates establish the Cohn-Elkies linear-programming bound cannot be sharp in dimensions 10 and 11 99%

Evidence

• huggingface.co https://huggingface.co/papers/2610.24944
• arxiv.org https://arxiv.org/abs/2610.24944

Actionability rated low.