Unverified50% confidenceBenchmarkExact time
qwen3.8 had a latency p50 of 18885ms and p95 of 42368ms, 15 to 20 times slower than Claude Sonnet 5's 1080ms median latency on the same task.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedOn GPU fp16 (RTX 5090), Largha-150M-v1 reports a p50 latency of 21.0 ms per decision.72% similarUnverifiedGPT-OSS-20B BF16 loading was reduced from 545 seconds to 70 seconds (7.8× speedup), and Qwen3.5-397B achieved a 1.93× to 2.3× improvement.70% similarVerifiedIn benchmark testing, Qwen3 27B achieved a generation speed of approximately 40-something tokens per second on 4 modified 2080 Ti cards without NVLink enabled.68% similarUnverified早期5.8GHz模拟图传延迟约15ms但画质仅480p，大疆O3/O4等数字图传将延迟压缩至20-30ms并将画质提升至1080p/60fps66% similarUnverifiedOn Qwen3.5-397B-A17B, TokenSpeed reported speeds of up to 580 tokens/second on agent workloads.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1047895API
curl https://kongchang.com/api/v1/knowledge/claims/1047895MCP
get_claim(id=1047895)