Fetching from the wire…
Public story · 2026-08-17 · high
The project publishes no benchmark scores, but a warm Lean REPL cuts each proof check from 30 seconds to 0.4 seconds.
Why now: MathCode's climb to 102 points on Hacker News put it in view on August 17.
MathCode keeps its Lean REPL warm, answering proof checks in about 0.4 seconds instead of 30, per the project's page. That's the gap between an automated prover trying a handful of strategies and one trying hundreds before a person stops waiting.
The tool takes a plain-language math problem, formalizes it as a Lean 4 theorem, and attempts a proof against reusable theorem and axiom libraries. An Obsidian view, the same app people use for note-taking, shows the resulting proof graph. Math-AI's page frames the REPL warmup as a one-time cost, after which checks stay fast for the rest of a session.
The code is open source at github.com/math-ai-org/mathcode. It reached 102 points on Hacker News, a sign people are curious even without a demonstrated proof rate. The page publishes no benchmark scores. It also doesn't say how often the plain-language formalization step lands on the correct Lean 4 statement before the proof search starts.
Each link below shares sources, entities, or timing with this story.
Claudian uses Obsidian / Shared entity: Obsidian / Earlier coverage
Linked by a graph relationship (Claudian uses Obsidian); both cover Obsidian; earlier Obsidian coverage from 2026-05-11.
Linked by a graph relationship (Claudian uses Obsidian); both cover Obsidian; earlier Obsidian coverage from 2026-03-03.
Terence Tao uses Lean / Shared entity: Lean / Shared topic / Earlier coverage
Linked by a graph relationship (Terence Tao uses Lean); both cover Lean; overlapping topics (attempt, capability).
OpenAI uses Lean / Shared entity: Lean / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (OpenAI uses Lean); both cover Lean; overlapping topics (attempt, check, math, proof).
Chad Arimura uses Obsidian / Shared entity: Obsidian / Earlier coverage
Linked by a graph relationship (Chad Arimura uses Obsidian); both cover Obsidian; earlier Obsidian coverage from 2026-08-13.
OpenAI uses Lean
Linked by a graph relationship (OpenAI uses Lean).
OpenAI uses Lean / Shared entity: Lean / Shared topic / Earlier coverage
Linked by a graph relationship (OpenAI uses Lean); both cover Lean; overlapping topics (claim, lean).
OpenAI uses Lean / Shared topic
Linked by a graph relationship (OpenAI uses Lean); overlapping topics (agentic, benchmark).