Unverified50% confidenceFactExact time
The benchmark run produced a total of 200 non-empty responses (100 each for classification and neighbor identification), and the task was fully executed.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/12/2026
First Seen
Valid until: 10/26/2026
Sources
Related Claims
UnverifiedCMU's Odyssey benchmark contains 200 tasks, each accompanied by multiple rubrics, supporting partial scoring rather than a single binary judgment.66% similarUnverifiedDuring the evaluation phase, the system solved 105 out of 120 tasks.64% similarUnverifiedIn autoregressive generation, producing a response containing 500 tokens requires 500 forward passes, which cannot be parallelized; whereas perception tasks such as image classification require only a single forward pass.64% similarUnverifiedThe research team manually annotated 100 real recording cases, labeling 348 scoring units and 162 localization events across ten combined label types.63% similarUnverifiedThe project has been selected for a relevant list of top 100 projects.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1007683API
curl https://kongchang.com/api/v1/knowledge/claims/1007683MCP
get_claim(id=1007683)