Legal Tasks: Models Meet 90% of Criteria but Fail Full Pass
ArtificialAnlys · x · 2026-07-08
Artificial Analysis found that models can handle most requirements of a legal task but rarely satisfy all of them. Leading models can achieve over 90% on individual scoring rubrics, yet their full pass rate for entire tasks sits between 0% and 14.2%. Only 4 models exceeded a 90% rubric pass rate: Fable 5 (93.6%, the highest), Opus 4.8 (91.1%), GLM-5.2 (91.0%), and Sonnet 5 (90.1%).
Related event: Artificial Analysis Releases Harvey Legal Agent Benchmark Results(8 posts)→
More from Models
- Same Echo Maze prompt, three frontier models: all passed visually but shipped the same hidden bug — eyishazyer · 2026-09-11
- Benchmark scores drop from 89% to 19% on new evals — how benchmaxxing breaks leaderboard trust — airesearch12 · 2026-09-11
- ChatGPT tells user their question is too hard and to 'accept dumber answers' — phido3000 · 2026-09-11
- Developer Building a Unified Leaderboard of All Model Benchmark Scores — airesearch12 · 2026-09-11
- Rumor claims Kimi faked performance by serving Claude; DeepSeek new model surprises in evals — realsohamparekh · 2026-09-11
- GPT-5.6 writes well but is instantly forgettable, user complains — BasedRaddka · 2026-09-11