Unverified50% confidenceBenchmarkExact time
Harvey and Engram used the model to build a synthetic law firm, achieving an average score of 67 across 250 legal tasks, higher than all models evaluated in the study.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/18/2026
First Seen
Valid until: 12/17/2026
Sources
Related Entities
Related Claims
UnverifiedA typical deep research Agent task involves 30–50 independent model inference calls, with cumulative token consumption easily exceeding 100,000.63% similarUnverified顶级AI模型（如Claude Opus）每次调用消耗的算力可能是轻量级模型的20-50倍60% similarUnverifiedIn the experiment where the 27B model wrote its own skills, it scored 56.3% on LiveMath.58% similarVerifiedIn 2022, Hoffmann et al.'s Chinchilla paper redid experiments on four hundred models and concluded that, under the premise of compute-optimality, each parameter should correspond to approximately 20 tokens.57% similarUnverifiedThe author found through statistics that there are more than 70 codebases following the "copy-and-hack" ACT or Diffusion Policy pattern.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/931007API
curl https://kongchang.com/api/v1/knowledge/claims/931007MCP
get_claim(id=931007)