Expired50% confidenceBenchmarkExact time
During the decode phase, Ornith was stable at 43 tokens per second, while Qwen's decoding speed was overall about 10 tokens per second slower.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/9/2026
First Seen
Valid until: 10/7/2026(expired)
Sources
16GB VRAM Showdown: Ornith 35B vs Qwen 35B — Full Benchmark Comparison
bilibili程序员-智能译站7/7/2026
Related Claims
UnverifiedDeploying AI inference capabilities to edge nodes can reduce latency from hundreds of milliseconds to single-digit milliseconds.61% similarUnverified只有在LLM推理比embedding加向量检索慢数十倍以上时，语义缓存才能同时实现降本与提速60% similarUnverified在GPU的SIMT执行模型下，同一warp中的32个线程需在同一时钟周期执行相同指令，Karatsuba的条件分支会引发线程分歧导致有效计算效率降至峰值的1/4甚至更低60% similarUnverifiedDoubao TTS exhibits an over-response problem in speech rate control, where synthesized output may become abnormally slow when slow speed and emphasis requirements are annotated simultaneously.59% similarUnverifiedOnce network latency exceeds 500ms, the fluidity of Cursor's code completions drops noticeably59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/342892API
curl https://kongchang.com/api/v1/knowledge/claims/342892MCP
get_claim(id=342892)