Unverified50% confidenceTradeoffExact time
The model's local throughput is approximately 9 to 22 tokens per second, consumes more tokens than the previous generation, and is slower when thinking mode is enabled.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/18/2026
First Seen
Valid until: 12/17/2026
Sources
Related Entities
Related Claims
Unverified高推理强度模式下单次请求的 Token 消耗可能是低强度模式的 5-20 倍69% similarUnverifiedThe newer the model, the better the results, but the more token quota it consumes.67% similarUnverified同一段文字在不同模型中消耗的 token 数量可能相差 10%-30%67% similarUnverifiedThe officially claimed 30% reduction in Thinking Token consumption only held true for two of the five problems; for the remaining three, K2.7-Code actually consumed more tokens.66% similarUnverifiedResearch indicates that when the context exceeds a certain length, the model's recall accuracy for content in the middle portion may drop below 50%.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/931008API
curl https://kongchang.com/api/v1/knowledge/claims/931008MCP
get_claim(id=931008)