Expired50% confidenceBenchmarkExact time
The measured inference speed of running a 90-million-parameter model on a PSP is approximately 0.5 to 0.6 tokens per second, with a complete response taking 1 to 3 minutes to generate.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
9/6/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 9/20/2026(expired)
Sources
Related Entities
Related Claims
UnverifiedThe inference latency of dedicated safety classification models can be kept within 200–500 milliseconds, and can be compressed to the 100-millisecond range with quantization techniques and GPU inference optimization.71% similarUnverifiedThe response time for a single inference request from a large model typically ranges from 1 to 30 seconds.69% similarUnverifiedGoogle's RAIL model indicates that users' perceptual threshold for response latency is approximately 100ms to 1000ms, and exceeding 3 seconds will most likely cause them to abandon waiting.68% similarUnverifiedSFT takes approximately 0.5 seconds per step, while DRAFT takes approximately 4.11 seconds per step, making DRAFT roughly 8 times slower per step than SFT.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/864619API
curl https://kongchang.com/api/v1/knowledge/claims/864619MCP
get_claim(id=864619)