Do Theorem Provers Need Theorem-Proving Agents? mini-Lean-agent: A Coding-Agent Baseline for Lean

Key points

• General-purpose coding agents use standard shell and file interactions, yet formal theorem-proving agents often rely on specialized, ad-hoc tooling.1
• We compare mini-Lean-agent against AxProverBase and Numina-Lean-Agent on the Putnam 2025 benchmark.1
• Experiments with Gemini 3.8 Flash as the base model show that mini-Lean-agent solves 7/12 problems at $2 per problem limit, compared with AxProverBase's 4/12.1
• Our results motivate using mini-Lean-agent as a standard baseline when evaluating theorem-proving infrastructure.1

Sources (1)

• [1]Do Theorem Provers Need Theorem-Proving Agents? mini-Lean-agent: A Coding-Agent Baseline for Lean
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 07:40 PM

  We introduce mini-Lean-agent, a deliberately minimal agent for Lean tasks whose only model-visible tool is bash.

  General-purpose coding agents use standard shell and file interactions, yet formal theorem-proving agents often rely on specialized, ad-hoc tooling.

Extractive summary: sentences quoted from the sources.