---
format: "aidr-story-markdown/v1"
id: "676eaec7e95ba659a52ddf8b95cd3dcba49f2e16da947fb7a35b77bda58bb8b4"
canonical_url: "https://aidr.today/676eaec7?lang=en"
title: "Mistral Open Model Agent Scores 82% on ArXivLean With 10 Million Tokens Per Problem"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-11T16:54:17.000Z"
category: "Research"
topics: ["mistral","kimi","theorem-proving","reasoning","benchmark","open-source"]
source_urls: ["https://huggingnews.com/ai/mistral-open-model-agent-scores-82percent-on-arxivlean-with-10-million-t-7b1f6b8e"]
summary: "A duo of open weights models was used to reach a new performance peak in automated theorem proving. The Mistral team employed Leanstral 1.5 and Kimi K3 to atta…"
---

# Mistral Open Model Agent Scores 82% on ArXivLean With 10 Million Tokens Per Problem

> [Open the canonical story](<https://aidr.today/676eaec7?lang=en>)

**Published:** 2026-09-11T16:54:17.000Z
**Category:** Research
**Topics:** mistral, kimi, theorem\-proving, reasoning, benchmark, open\-source

## Summary

A duo of open weights models was used to reach a new performance peak in automated theorem proving\. The Mistral team employed Leanstral 1\.5 and Kimi K3 to atta…

## Sources

- [Story source](<https://huggingnews.com/ai/mistral-open-model-agent-scores-82percent-on-arxivlean-with-10-million-t-7b1f6b8e>)