micro1 launches LegalReasonBench to test models on attorney-authored legal matters
douwekiela · x · 2026-10-17
micro1 introduces LegalReasonBench, a new benchmark testing whether AI models can handle the complex multi-step, long-horizon reasoning of real legal matters, with tasks authored by practicing attorneys. CEO Douwe Kiela frames legal reasoning — distinctly human and deeply nuanced — as "one of the ultimate stress tests of AGI," aligning with the company's focus on knowledge-work capabilities.
More from Research
- visloc-rs: pure-Rust SfM 3.46x faster than COLMAP, beats ORB-SLAM3 on 9/11 EuRoC — rsasaki0109 · 2026-10-17
- AgentForesight: a 7B online auditor that catches multi-agent failures mid-run — jiqizhixin · 2026-10-17
- Yann LeCun and NYU team propose AdaJEPA to adapt world models when the world changes — mengyer · 2026-10-17
- Harvard mathematician open-sources attempt to get Claude to construct Yang-Mills — lvwerra · 2026-10-17
- Model tweaks barely move MMLU but easily break generative tasks — aryaman2020 · 2026-10-17
- Idea: Give Agents a 'Fluid Dynamics Lab' to Run Experiments Autonomously — OA_paperclips · 2026-10-17