Unverified50% confidenceBenchmarkExact time
Experiments found that the task_score under the shaped reward configuration was only 0.125, compared to 0.417 for the naive reward configuration, with an effect size of d=4.47.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/7/2026
First Seen
Valid until: 12/6/2026
Sources
Related Entities
Related Claims
UnverifiedOSWorld 2.0采用细粒度检查点部分奖励评分，平均每个任务有27.25个检查点，不要求固定顺序66% similarUnverifiedIn the experiment, the 27B model scored 61.0% on LiveMath when skills were written by the 4B model, whereas it scored 56.3% when it wrote the skills itself63% similarUnverified跑分排名不等于所有场景的实际表现，Terra跑分接近5.5但游戏任务中5.5完成度略高63% similarUnverifiedMiniMax M2.7 asks 0.6 questions per task, but its clarification advancement rate is only 60.7%, indicating a problem of low-information-gain questioning62% similarUnverifiedSonnet 5's Max Effort mode produces approximately 40% more output tokens per Intelligence Index task compared to Sonnet 4.6.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/871626API
curl https://kongchang.com/api/v1/knowledge/claims/871626MCP
get_claim(id=871626)