Unverified50% confidenceOpinionExact time
The evaluation methods of leaderboards like Vals Index mostly focus on short outputs or medium-length outputs, with a single output rarely exceeding a few thousand tokens, so the capability for million-token outputs is hard to directly reflect in leaderboard scores.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/14/2026
First Seen
Valid until: 1/12/2027
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66% similarUnverifiedPreviously, the output limits of mainstream models mostly stayed in the range of tens of thousands to just over a hundred thousand tokens.61% similarUnverified利用Row Group的min/max统计元数据进行谓词下推，在选择性高的查询中可跳过90%以上的数据读取59% similarUnverified即使模型标称支持128K甚至1M token的上下文窗口，其有效利用能力往往远低于标称值，实际可靠工作范围可能只有标称值的几分之一58% similarUnverifiedIn many evaluations conducted on small datasets, scores fluctuate by several percentage points across different runs.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1024315API
curl https://kongchang.com/api/v1/knowledge/claims/1024315MCP
get_claim(id=1024315)