Unverified50% confidenceBenchmarkExact time
在Gemma 4 12B模型的Token生成测试中，Spark达26.4 tokens/s、Halo 24.6 tokens/s、M4 Pro最快33.8 tokens/s
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/12/2026
First Seen
Valid until: 10/10/2026
Sources
AMD Ryzen AI Halo实测：x86架构能否撼动英伟达DGX Spark？
bilibili量子菠萝_7/7/2026
Related Claims
Unverified根据Artificial Analysis数据，Fable完成单个任务约用33k token，Opus 5约用37k token73% similarUnverifiedChinchilla模型拥有70B参数和1.4T tokens，在同等算力下显著优于拥有280B参数和300B tokens的Gopher模型70% similarUnverifiedMiniMax M3 每个 token 推理时约激活 23B 参数69% similarUnverifiedIn the e-commerce website test, Opus 4.8 consumed 4.898 million output tokens while Fable 5 used only 588K tokens66% similarUnverified父子分块策略用小块（128~256 Token）做精准检索，命中后返回其所属大块（512~1024 Token）作为上下文66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/489929API
curl https://kongchang.com/api/v1/knowledge/claims/489929MCP
get_claim(id=489929)