Unverified50% confidenceBenchmarkExact time
The five representative memory systems tested achieved an aggregate score of at most 45% across all dimensions.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/17/2026
First Seen
Valid until: 2/15/2027
Sources
Related Entities
Related Claims
UnverifiedIn 4,580 controlled integration experiments conducted on the gpt-oss-120b endpoint (stratified sampling from 458 committees), the sample estimate was -0.09 percentage points, and the design-weighted gain was +0.94 percentage points (95% confidence interval [-1.31, +3.50]), while reducing input by 18.72%.71% similarVerifiedExperiments by the Long-Term team showed that using the same model with only adjustments to rule constraints and the memory system raised the score from 52.8% to 66.5%, jumping the ranking from Top 30 to Top 5.71% similarUnverifiedCompared with the best-performing simple generator, the model leads on 5 of 6 items on the main test set, with advantages of 0.7 to 2.5 percentage points, and the test can prove the gap for 1 of those items.68% similarUnverifiedFor an unconstrained all-layer adapter, 21% to 51% of its gains are routed through the attention of the same set of key heads, compared with at most 7% for random heads.68% similarUnverifiedMedIQ多智能体系统最终达到57%准确率,摘要智能体带来约5%提升,而拥有全部前置信息的Oracle上限为66%68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1179635API
curl https://kongchang.com/api/v1/knowledge/claims/1179635MCP
get_claim(id=1179635)