Unverified50% confidenceBenchmarkExact time
该配置下35B模型实测单人最高速度约80 Tokens/秒，并发可达500 Tokens/秒以上
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/11/2026
First Seen
Valid until: 12/10/2026
Sources
Related Entities
Related Claims
Unverified在对照测试中，35B模型速度约100 tokens/秒（Mac Studio），是9B模型16 tokens/秒的约五倍79% similarUnverified在4×RTX 3090 SXM4配置下，GLM-5.2量化版本的Token生成速度约为6.3-6.8 tokens/秒，提示评估速度约为10.6 tokens/秒77% similarUnverified在预填充速度测试中，Halo约650 tokens/s，Spark约2000 tokens/s，Spark快了将近三倍77% similarUnverified本地部署DS4模型在STM32嵌入式开发测试中平均输出速度约为23 tokens/秒，推理时内存占用飙升至110GB，峰值速度达83.8 tokens/秒76% similarUnverified基于MLX框架，顶配Mac机型运行Llama-3-70B量化模型可达约20 tokens/秒的生成速度76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/902433API
curl https://kongchang.com/api/v1/knowledge/claims/902433MCP
get_claim(id=902433)