Unverified50% confidenceFactExact time
Doubling the thinking token budget from 8k to 16k brought only a 0.2% accuracy improvement, while changing the distribution of failed questions.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
Unverified研究数据显示，在复杂代码生成任务中，将thinking budget从1024 tokens提升到8192 tokens可以带来15-30%的准确率提升，但推理成本增加4-8倍77% similarUnverifiedThis approach reduces token usage by 58.3% and improves inference speed by 1.95×, with no loss in accuracy.73% similarUnverifiedA lower cost per token does not mean a lower cost per correct decision; once the cost of judgment errors is factored in, a $0.02 model with a 15% error rate may be more expensive than a $0.10 model with a 5% error rate.71% similarUnverifiedToken consumption for reasoning at the XHigh setting can reach 3–5 times that of the High setting, but the marginal accuracy improvement is typically only 5–15 percentage points.69% similarUnverifiedOn the arithmetic control task, all models' accuracy stayed the same but the cost per correct answer rose by 1.65 to 2.5 times.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1038587API
curl https://kongchang.com/api/v1/knowledge/claims/1038587MCP
get_claim(id=1038587)