LLM Evaluation & Leaderboards
News on LLM evaluation and benchmarks, covering new tests, methodology changes, leaderboard results, and whether findings can be reproduced.
Explore how researchers and engineering teams test language models before and after deployment. Coverage looks at benchmark design, meaningful comparisons, leaderboard changes, and which results teams can reproduce.
Latest coverage
Articles
NewsMBZUAI details CT-SpatialVQA; eight tested models score below 50%
NewsMIT benchmark says GPT-5.5 matched 34.6% of recent theory problems
NewsApple study finds tool-discovery gains shift with task complexity
NewsCohere tests the cost-performance trade-off in coding-agent teams
NewsAnthropic cuts live internet access from all internal evaluations
NewsScale Labs and Elorian release Humanity’s Sixth Sense benchmark
NewsGoogle trial finds AI improves patent drafts, but junior judgment does not improve on average
NewsFudan researchers reconstruct part of OpenAI Navier–Stokes work, defer key step
NewsScale AI releases SWE-Bench Pro V2 with locked evaluations
NewsMicrosoft releases ThinkingBox agent benchmark through Hugging Face
NewsMIT and Sakana AI propose SIFT to prioritize coding-agent benchmark runs
NewsApple researchers introduce SCLATE for continual-learning agent evaluation