Unverified50% confidenceBenchmarkExact time
Google reports that on MoE models of 235B to 480B parameters, KV transfer latency was reduced by more than 90%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/13/2026
First Seen
Valid until: 2/11/2027
Sources
Related Entities
Related Claims
UnverifiedGPT-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.62% similarUnverifiedGoing from 16G to 24G of VRAM, median decoding speed jumps from 56.8 tok/s to 98.3 tok/s, a 73% increase.61% similarUnverifiedAdding a third card to a dual-card setup increased decoding by only 1% or even had a negative effect; 128K domain fill speed dropped from 57 to 22.95 tok/s.61% similarUnverifiedCompared with the full-rank (no low-rank approximation) approach, VPU at dim=256 reduced 200 steps from 4249ms to 160ms, achieving a 26.44x speedup.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1166530API
curl https://kongchang.com/api/v1/knowledge/claims/1166530MCP
get_claim(id=1166530)