Unverified50% confidenceBenchmarkExact time
In this experiment, the inference latency for each decision step of Kolibri-1 was kept within 25 milliseconds.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/7/2026
First Seen
Valid until: 1/5/2027
Sources
Related Entities
Related Claims
UnverifiedDirectly reading the probability values from the logits layer, bypassing the redundant text generation step, is the key to achieving the low latency of 25 milliseconds.69% similarUnverifiedWait-k策略中k值选择在延迟与准确率之间形成权衡，k=0时延迟最低但准确率最差，k趋近句子长度时退化为离线翻译68% similarUnverifiedIn 1993, Jakob Nielsen proposed the three-threshold theory of human-computer interaction latency: 0.1 seconds for instant feedback, 1 second before the flow of thought is interrupted, and 10 seconds as the limit for maintaining attention.66% similarUnverified在每个动作前进行策略校验可能给智能体响应速度带来延迟，这是所有运行时安全方案都需面对的经典权衡66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/984860API
curl https://kongchang.com/api/v1/knowledge/claims/984860MCP
get_claim(id=984860)