Lean 4

2026-10-31

RESEARCH · 1 SOURCE · arXiv cs.AI

OpenFrontiers: dataset and benchmark for LLM scientific rediscovery (arXiv:2610.37739v1)

OpenFrontiers is a cross-disciplinary resource combining a structured corpus of 12,151 open research problems with a retrospective benchmark of 200 mathematics problems (110 resolved in 2026, 90 still open), each formalized as Lean 4 statements with automated checks and independent reviews; a higher-quality HQ-50 subset is provided. The authors publish an evaluation protocol (closed-book, frozen-retrieval, live-web) and report results evaluating seven LLM agents on HQ-50 under a standardized budget, finding closed-book rediscovery rare (at most 7 of 40 resolved problems) while live-web successes largely reflect retrieval of published solutions; the dataset is hosted at the linked Hugging Face collection.

7.0

2026-10-23

RESEARCH · 1 SOURCE · arXiv cs.AI

PerMIT: machine-checked integrity theorems and a trust-gated memory architecture for LLM agents

PerMIT is a memory architecture for LLM agents that enforces a strict trust model—only user-typed inputs and values confirmed at a gate are trusted—and mediates untrusted environment-derived values via a finite interface. The authors prove two integrity theorems (machine-checked in Lean 4), report zero observed attack success on Trojan Hippo with self- and lightweight verifiers, and simulate communities (up to 90 agents) on τ^2-bench showing scalable community-confirmation of memory and improved downstream agent performance.

7.0

2026-10-21

RESEARCH · 1 SOURCE · arXiv cs.AI

Bounded compiler-guided repair improves tactic-level proof search in Lean 4

The paper introduces a bounded, online step-level repair that samples K candidate tactics per proof step and, when a tactic is rejected, feeds the compiler error message to a pretrained repair model to fix that step in place up to R times. On a 1,990-theorem subset of the LeanDojo benchmark the method yields up to a 32% relative improvement (82 additional theorems solved) across tested K∈{1,3,5,8}, replicates on the Lean Workbook benchmark (nearly doubling solve rate), and shows gains concentrate on shallow tactic-execution failures at the expense of predictable extra runtime.

6.0

2026-10-21

RESEARCH · 1 SOURCE · arXiv cs.AI

ReuF2F: Lean 4 benchmark of 244 REU open math problems for coding agents

ReuF2F is a new Lean 4 benchmark (arXiv:2610.27046v1) of 244 formalized open problems sourced from Research Experiences for Undergraduates (REU). The benchmark packages each problem and its negation in a Lean project inside a Docker container, provides a unified evaluation harness (Leanimum-agent built on mini-SWE-agent), and reports that the best model in their evaluation solves only ~40% of the problems.

7.0

2026-03-16

MODELS · 1 SOURCE · Mistral AI

Leanstral: open-source Lean 4 proof-engineering agent (Leanstral-120B-A6B) released under Apache 2.0

Leanstral, an open-source code agent trained for Lean 4 proof engineering and shipped as Leanstral-120B-A6B, is released with Apache 2.0 weights, an agent mode for Mistral vibe, a free API endpoint, a forthcoming tech report, and a new evaluation suite called FLTEval. The model uses a highly sparse architecture (6B active parameters), supports MCPs via vibe (optimized for lean-lsp-mcp), and is benchmarked on realistic repository tasks where it shows cost and efficiency advantages versus several closed- and open-source competitors.

8.0