Unverified85% confidenceFactTime unknown
Sonnet 4.6单独执行在SWE-Bench上得分72.1，成本$1.09；而Sonnet 4.6执行加Opus顾问得分74.8，成本$0.96
1
Sources
85%
Confidence
Medium-term (~90 days)
Relevance
6/1/2026
First Seen
Valid until: 8/30/2026
Sources
Claude顾问策略：小模型执行大模型把关，AI Agent省钱提效新范式
bilibili灵姐说AI
Related Entities
Related Claims
Unverified在SWE Bench Pro编程基准测试中，Sonnet 5取得63.2%，Opus 4.8取得69.2%81% similarVerifiedSonnet 5在Terminal Bench 2.1上取得80.4%的成绩，接近Opus 4.879% similarUnverifiedSonnet 5相较Opus 4.8 Max仅便宜约72%75% similarUnverified在SWE-Bench Pro基准上，Sonnet 5明显落后于Opus 4.874% similarPartially Verified官方称Sonnet 5在推理、工具使用、编码和知识工作方面相比Sonnet 4.6均有显著提升，整体性能已逼近OPUS 4.8但价格更低74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/27289API
curl https://kongchang.com/api/v1/knowledge/claims/27289MCP
get_claim(id=27289)