<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Numina-Lean-Agent: An Open and General Agentic Reasoning System for Formal Mathematics · HackerPapers discussions</title><link>https://hackerpaper.kaipfister.de/p/2601.14027/</link><description>Collected HN comments and submissions linking this paper.</description><language>en</language><lastBuildDate>Mon, 02 Nov 2026 11:27:26 GMT</lastBuildDate><atom:link href="https://hackerpaper.kaipfister.de/p/2601.14027/feed.xml" rel="self" type="application/rss+xml" /><item><title>Numina-Lean-Agent: An Open and General Agentic Reasoning System for Formal Mathematics — HN submission</title><link>https://news.ycombinator.com/item?id=49010310</link><guid isPermaLink="false">urn:hackerpapers:submission:2601.14027:49010310</guid><pubDate>Wed, 22 Jul 2026 17:28:39 GMT</pubDate><description>&lt;p&gt;Grok is a surprisingly good automated theorem prover&lt;/p&gt;&lt;p&gt;TL;DR: I&amp;#x27;m working on a Python package called OpenATP [1] that provides a common interface to coding agents for automated theorem proving in Lean. In the latest release, I added support for Leanstral 1.5 [2,3], Grok, and Kimi Code [4]. I was surprised to find that Grok has accuracy competitive with Claude Code and Codex at faster wall-clock times and a fraction of the cost. See the rest of this post and the OpenATP docs [5] for more details. Automated theorem provers take formal statements (in a proof assistant like Lean) and attempt to synthesize a proof. Unsurprisingly, LLMs have become the…&lt;/p&gt;&lt;p&gt;3 points; 1 discussion reply on HN. &lt;a href="https://hackerpaper.kaipfister.de/p/2601.14027/"&gt;Paper and source history&lt;/a&gt;. A submission is not an endorsement.&lt;/p&gt;</description><category>cs.AI</category></item></channel></rss>
