Unverified60% confidenceFactExact time

CoT效果在参数量超过约100B的模型上尤为显著，体现为「涌现能力」

2

Sources

60%

Confidence

Long-term

Relevance

7/18/2026

First Seen

Sources

Opus 5传闻发酵：Anthropic的用户信任危机与产品迭代争议

Related Claims

Unverified研究发现CoT推理能力只在参数量超过约100B的超大规模模型中自发涌现81% similar Verified思维链效果在参数量超过100B的模型上尤为显著74% similar VerifiedCoT的涌现性表现为仅当模型参数量超过约1000亿时才能显著提升性能，小模型上几乎无效72% similar Unverified上下文控制在100K token以内时模型表现最为稳定，超过100K后幻觉率明显上升65% similar Unverified研究表明CoT提示在参数量低于100B的模型上几乎无效64% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/549729

API

curl https://kongchang.com/api/v1/knowledge/claims/549729

MCP

get_claim(id=549729)