Unverified50% confidenceFactExact time
When the context extends to 100k or even longer, the per-token price, memory usage, and inference latency all change, and "performance per dollar" conclusions under short contexts cannot be directly extrapolated.
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedReasoning tokens do not appear in the final response, but they count toward usage and cost while also slowing down response speed.74% similarUnverifiedShorter output tokens can simultaneously slow context expansion and reduce per-step generation time, making long loops more affordable in both cost and latency.70% similarUnverified推理模式的token消耗通常是普通模式的3-10倍，thinking tokens生成通常增加5-30秒延迟69% similarUnverifiedFor large models, "inference efficiency" usually refers to the number of tokens that can be generated per unit time (throughput) and the first-token latency.68% similarUnverified长程任务的Token消耗随步骤数呈近似二次方增长，而非线性增长67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1102879API
curl https://kongchang.com/api/v1/knowledge/claims/1102879MCP
get_claim(id=1102879)