Unverified50% confidenceFactExact time
LeNet-5约有7层，参数量小
1
Sources
50%
Confidence
Long-term
Relevance
8/16/2026
First Seen
Sources
Related Claims
UnverifiedOn the SWE-Bench Pro programming benchmark, Claude Fable 5 leads with a score of 80%, ahead of GPT-5.6 Sol's 64.6%.62% similarUnverified在MoE模型中每个token只激活少量Expert（通常2-4个，Top-K路由），实际计算量可能仅相当于6-7B的Dense模型60% similarUnverifiedIn the initial publicly released SWE-bench results, GPT-4's resolution rate was only approximately 1.7%, Devin claimed to reach 13.86%, and SWE-agent reached 12.5%.58% similarUnverifiedThe median output token count per problem-solving attempt for Opus 4.7 is 60,000, while GPT 5.5 requires only 16,000.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/757792API
curl https://kongchang.com/api/v1/knowledge/claims/757792MCP
get_claim(id=757792)