Unverified50% confidenceFactExact time
An analysis by a Princeton team of 9 models, 9 benchmarks and 21,730 runs reported that scaffolds significantly change both cost and accuracy; among 36 model-agent-benchmark combinations, 21 showed accuracy declining or staying flat after increasing reasoning intensity.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
AI Agent Harness Explained: How 66 Frameworks Are Reshaping Agent Capabilities
youtubeAI Briefing Daily11/3/2026
Related Entities
Related Claims
UnverifiedThe paper reports that hybrid reasoning agents achieve accuracy about 10 percentage points higher than conventional models on multiple test sets.74% similarUnverifiedIn a simulation of 2000 requests, the small model (1 unit cost/request) had an accuracy of 74.5% and a cost of 1.34 per correct answer; the large model (20 unit cost/request) had an accuracy of 95.8% and a cost of 20.88 per correct answer.73% similarUnverifiedIn the configuration using ten policy snapshots, using only 8 prompts achieved an average accuracy of 24.09%, approaching the 24.51% achieved with 14,080 prompts.73% similarUnverified在工具数量超过20个的场景下，工具描述的语义歧义度与模型选择准确率之间存在显著的负相关关系73% similarUnverified近期研究表明即便是参数量超过万亿的模型，在多位数乘法上的准确率也会随位数增加而指数级下降72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1135174API
curl https://kongchang.com/api/v1/knowledge/claims/1135174MCP
get_claim(id=1135174)