Aletheia math research agent (Gemini Deep Think)
Aletheia, a Deep Think math agent with a natural-language verifier, produced a research paper with no human input and solved four open Erdős-database questions.
Generate-verify-revise loop that can admit failure and uses search to avoid spurious citations. A semi-autonomous sweep of 700 open Erdős problems yielded four autonomous solutions; Deep Think scored up to 90% on IMO-ProofBench Advanced as inference compute scaled.
- Date
- Wednesday, 11 February 2026
- Lab
- Google DeepMind
- Kind
- paper
- Access
- research preview
Figures
| Measure | Value | Measured by |
|---|---|---|
| IMO-ProofBench Advanced | up to 90% Deep Think, scales with inference compute; graded by human experts | company |
Google proposes a taxonomy of AI-assisted math results and claims no Level 3 or 4 results; Level 2 "publishable quality" papers were submitted to journals. Page dated 2026-02-11 (RSS lists Feb 9).
Sources
This record was checked against its sources on 6 October 2026. How we check