Unverified50% confidenceFactExact time
LARY-Bench is a benchmark for evaluating the quality of latent action representations, using whether actions are correctly encoded and whether they can be reused across samples as evaluation dimensions, rather than simply looking at reconstructed pixel similarity.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe researchers built ReGap-Bench, a paired benchmark containing five types of visual reasoning tasks, to test the effect of external reference on model performance.73% similarUnverifiedTrajectory quality refers to whether the series of intermediate actions of the agent between receiving the query and giving the answer is reasonable, including the order of tool calls, whether retrieval parameters are on topic, whether invalid searches are repeated, and whether verification continues when evidence is insufficient.71% similarUnverifiedThe evaluation scope of TrialPD-Bench is not limited to final answer correctness, but also includes the evidence recall rate of agent trajectories, agent behavior patterns, and cost.71% similarUnverifiedEach sample in FastBench comes with a manually annotated evidence interval, and the evaluation not only judges whether the answer is correct but also can locate whether the model attended to the correct time segment.70% similarVerifiedA benchmark is a fixed set of questions used to score and horizontally compare models, similar to a mock exam, and is used to rank models.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1134626API
curl https://kongchang.com/api/v1/knowledge/claims/1134626MCP
get_claim(id=1134626)