Unverified50% confidenceReleaseExact time
8月13日OpenAI预览了Ultrafast，一个全新API层级，运行旗舰模型速度最高可达常规14倍，约每秒输出750个Token，运行在Cerebras硬件上
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/18/2026
First Seen
Valid until: 11/16/2026
Sources
Related Claims
Unverified旗舰模型Sol运行于Cerebras晶圆级芯片上，推理速度达750 Token/秒，约为上一代的15倍69% similarUnverifiedFastAPI的单机并发吞吐量可达传统同步框架（如Flask）的数倍至十倍以上68% similarUnverifiedSo将在Cerebras平台上线，推理速度可达每秒约750个Token，约为传统GPU集群的3-5倍65% similarUnverified该GDScript+Vulkan方案的运行速度约为llama.cpp+CUDA方案的1/1062% similarUnverified开源物理模拟库RocketSim能以约100倍实时速度在CPU上复现《火箭联盟》的物理行为62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/771251API
curl https://kongchang.com/api/v1/knowledge/claims/771251MCP
get_claim(id=771251)