Unverified50% confidenceFactExact time
苹果A18 Pro芯片的Neural Engine峰值算力达到约35 TOPS，足以本地运行30亿至70亿参数量级的压缩语言模型
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedApple Silicon Neural Engine的峰值算力从M1的11 TOPS跃升至M4的38 TOPS，性能提升超过3倍73% similarUnverifiedThe Qualcomm Snapdragon 8 Elite's NPU peak compute is approximately 45 TOPS, the Apple A18 Pro's Neural Engine is 35 TOPS, and the RTX 4090's peak compute exceeds 1,300 TOPS.66% similarUnverifiedThe Raspberry Pi 5 paired with quantized models can run language models at the 3B parameter level, while the NVIDIA Jetson Orin series with up to 275 TOPS of AI compute can run 7B or even 13B quantized models with latency controllable within 1 second64% similarUnverified根据Cursor自研的CursorBench基准测试，Opus 4.8相比前代Opus 4.7在编码效率上有显著提升63% similarUnverifiedxAI的Grok-1采用3140亿总参数、每次激活约25%专家的设计62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/566527API
curl https://kongchang.com/api/v1/knowledge/claims/566527MCP
get_claim(id=566527)