Unverified50% confidenceBenchmarkExact time
On the 23 FDE-Gym tasks where all key facts had been obtained, the best configuration achieved meaningful business improvement on only 10 tasks.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/19/2026
First Seen
Valid until: 1/17/2027
Sources
Related Entities
Related Claims
UnverifiedOn FDE-Gym, the best configuration achieved an average score of only 0.56, solving only 14 of the 95 tasks.71% similarUnverifiedIn the RULER-16K experiments, KVFetch improved the average score across 13 tasks by +8.4, with gains concentrated on tasks requiring sequential access.61% similarUnverifiedA total of 23 models received scores on both the resolve and resolve-plus tasks.59% similarUnverifiedIn a small-scale test published by the PartsPlan developer, all 30 chart values were answered correctly.56% similarUnverifiedResearch shows that 30B-parameter models have a success rate 15–30 percentage points lower than GPT-4/Claude-3.5 on complex reasoning tasks.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1066085API
curl https://kongchang.com/api/v1/knowledge/claims/1066085MCP
get_claim(id=1066085)