OEIS Open ($50)
Math · 2026-08-12
492 open OEIS conjectures formalized in Lean. Epoch's basic Inspect ReAct agent attempts each proof or disproof with a $50 budget; the original release uses SafeVerify in isolated compilation and scoring containers. Kept distinct from subsequent Comparator re-verification.
Top models (higher is better)
| Model | Score |
|---|---|
| Opus 4.8 | 29.9 |
| GPT-5.5 | 26.2 |
| Gemini 3.5 Flash | 22.2 |