Benefit Ledger · provisional · Mathematics
On 11 February 2026 DeepMind reported that Aletheia, a math agent on Gemini Deep Think, autonomously solved Erdős 1051 (motivating a generalization), disproved clauses of Erdős 1040 and a strong pinned-distance form of 654, and generated a publishable arithmetic-geometry paper on eigenweights without human intervention.
Current score
+0.27
Autonomous publishable math and several Erdős-database solves are notable (tier 2); DeepMind itself does not claim a landmark. Attribution to Deep Think / Aletheia (0.75). Evidence is a lab blog plus problem-database updates (0.40). Realization is posted research, some submitted (0.50).
This is one announcement, distinct from the already-scored IMO 2025 gold Deep Think event. DeepMind classifies the work as Level 1–2 (publishable), not Level 3–4 landmark. Four Erdős-database items and the eigenweights paper share the agent, model, and date, so they are one ledger event. IMO gold is not re-imported.
Reasoning mode inside the Aletheia math research agent that produced the Erdős solves and the eigenweights paper.
DeepMind attributes the February 2026 math-agent results to Gemini Deep Think mode, the same named family as the IMO 2025 gold system but a later research deployment.
Attribution 0.7500 · Credit share 100% · Google DeepMind
Aletheia autonomously solved Erdős 1051 and contributed constructions on 1040 and 654, and generated an eigenweights paper without human intervention.
outcome · supported
DeepMind classified these as landmark (Level 3–4) breakthroughs.
significance · disputed
primary sources
Google DeepMind · developer announcement · 11 Feb 2026
independent sources
erdosproblems.com · other · 11 Feb 2026
Secondary domains: Computer Science
13 Sep 2026 · 0.00 → 0.27
Imported events under methodology 0.1.
