GPT Codex Multi-Agent System Seals Over 15,000 Theorems in Lean
GiorgioPatrini · x · 2026-08-07
A tweet demonstrates the cutting-edge progress of automated theorem proving in 2026, highlighting high automation in AI mathematical reasoning.
- Multi-Agent Collaboration: The developer built a harness using Codex GPT Sol Ultra. The main agent (S0) runs computations on a cluster server and communicates with the Fable agent (L) via a mailbox mechanism.
- Autonomous Task Delegation: To accelerate progress, the system autonomously generated instructions to spawn multiple new agents (S1, S2, S3) on a home server to share the workload.
- Results: Running for 3 weeks, the project has successfully proved about 15,300 theorems in Lean, with only 4% of the project remaining.
More from coding & agent
- Warp Launches Agent CLI to Reimagine Subagent Orchestration UX — vikvang1 · 2026-08-07
- Scaling AI Agent Capacity 30x with a Fast Resumable Stateful Sandbox — rakyll · 2026-08-07
- Hermes Agent Adds Local Parsing for All Document Formats — Teknium · 2026-08-07
- Dev shows ultra-fast resumable stateful sandbox for tool calls — rakyll · 2026-08-07
- Agent Plugin Spec: Standardizing Packaging While Leaving Trust Decisions to Clients — Particular_Luck80 · 2026-08-07
- Asari Co-Inventor Agents Boost Kimi K3 Inference Speed by 32% — yisongyue · 2026-08-07