Unverified50% confidenceFactExact time
FINAL Bench is a functional metacognition evaluation benchmark covering 15 domains and 100 tasks, aimed at assessing models' ability to identify their own errors and correct them.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFINAL-Bench is the framework and competition infrastructure of this benchmark, responsible for defining evaluation criteria, the problem set, and the leaderboard mechanism, with a role similar to an evaluation protocol layer.73% similarUnverifiedThe WorkspaceBench benchmark contains 27 evaluation families with a total of 3356 questions, covering topics such as safety, logical reasoning, and multi-hop computation.68% similarUnverifiedTerminal Bench tests what a model can do rather than what it knows, measuring the model's complete closed-loop capability of planning, executing, observing feedback, and correcting errors66% similarUnverifiedUC-Bench is a human-annotated benchmark specifically designed to evaluate the ability to detect user-side conflicts.66% similarUnverifiedStructTokenBench is an evaluation benchmark suite for protein structure tokenizers, covering 7 benchmarks, 14 tasks and 28 test splits, with evaluation dimensions including structure reconstruction, function prediction, retrieval and others.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1058221API
curl https://kongchang.com/api/v1/knowledge/claims/1058221MCP
get_claim(id=1058221)