Reproduce “Mistral Large 4” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mistral AI, Abdelaziz Bounhar, Abhav Mehrotra, Abhijeet Somani, Aditi Kabra, Adrian Valente, Adrien Petralia, Adrien Sadé and 380 more
Mistral Large 4 is a multimodal mixture-of-experts model whose authors examine long-context attention for agentic coding. They find architecture variants with similar language-modeling loss can differ on SWE-bench Verified, while knowledge-based evaluations show no analogous gap. The final system restores long-context access in more layers and uses DSA to select past tokens; its benchmark results do not isolate DSA’s effect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.