Unverified50% confidenceBenchmarkExact time
In this configuration, Qwen3 27B achieves approximately 70 T/s natively in llama.cpp, with a measured peak of 76 T/s.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/1/2026
First Seen
Valid until: 11/30/2026
Sources
Related Entities
Related Claims
UnverifiedCalling Qwen3 27B within an Agent framework (Hornis/赫米) can maintain approximately 60 T/s.73% similarUnverifiedIn an RTX 5070 (12GB VRAM) configuration, the local Qwen3 9B model's output speed reached approximately 70–76 tokens/second.69% similarUnverifiedWhen deploying Qwen3 27B with 16 GB of VRAM, if some weights are offloaded to system memory, the speed drops sharply from tens of T/s to single digits.65% similarUnverifiedQwen3.6 27B supports a context length of 262K, and the 34B 80-layer version maintains the same context length unchanged64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/837542API
curl https://kongchang.com/api/v1/knowledge/claims/837542MCP
get_claim(id=837542)