Unverified50% confidenceFactExact time
ARC-Bench provides a leak-free, fixed-candidate test protocol that directly measures whether a frozen JEPA-style objective function can correctly rank candidate actions.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/11/2026
First Seen
Valid until: 12/10/2026
Sources
Related Entities
Related Claims
UnverifiedThe research team developed the ARC-Bench benchmark specifically to audit whether JEPA-style world models correctly rank candidate actions.79% similarUnverifiedBenchmark evaluation methods combine automated metrics (CLIP-Score, LPIPS, SSIM) with human evaluation.66% similarUnverifiedThis evaluation framework includes five benchmarks: ARC-AGI-V2, SWE-Bench, Terminal Bench 2.0, GPQA, and MMLU.65% similarUnverifiedAgent evaluation benchmark suites include GAIA, AgentBench, and SWE-bench.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/895201API
curl https://kongchang.com/api/v1/knowledge/claims/895201MCP
get_claim(id=895201)