Unverified50% confidenceFactExact time
KEEP-Bench contains 4,102 cases covering three challenging scenarios: retaining evidence to support plausible referential resolution, explicit qualifiers, and minority viewpoints.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedEven with customized prompts, the strongest models' evidence retention on KEEP-Bench remains unreliable.68% similarUnverifiedOn KEEP-Bench, the best closed-source model retained all required evidence in only 61.6% of cases.66% similarUnverifiedThe Evidence Plane allocates evidence budgets according to query needs, including distributing bounded evidence budgets among complementary evidence packets, allowing limited additional support, and rendering retained assertions into answer-ready context.61% similarUnverifiedA manual analysis of 250 error cases by the research team revealed that retrieval failure remains the primary bottleneck, with incorrect entity evidence being a particularly prominent issue.61% similarUnverifiedThis benchmark has the following limitations: judge reliability issues, the sample size limitation of each of the 100 scenarios being run only once, a single pressure condition, a pure text environment, the <think> block not guaranteeing access to the model's complete internal reasoning, and limited generalization ability.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1073295API
curl https://kongchang.com/api/v1/knowledge/claims/1073295MCP
get_claim(id=1073295)