Unverified50% confidenceBenchmarkExact time
StepBack was evaluated on seven mathematical reasoning and multi-hop QA benchmarks, and the results show that average accuracy is comparable to the self-consistency method while token usage is reduced by up to 41.26%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/6/2026
First Seen
Valid until: 2/4/2027
Sources
Related Entities
Related Claims
UnverifiedWhen using the multi-agent verification process, MARCH's accuracy was suppressed to 4.4%, whereas when the same model was directly asked which solution was better, accuracy reached 43.7%.75% similarUnverifiedBy selectively editing the expert routing scores to suppress overthinking, the median number of generated tokens was reduced by 30.4% and by up to 63.3% across four models and four benchmarks.73% similarVerifiedResearch indicates that at Q4 precision and above, performance loss is within 1–3% for most tasks.73% similarUnverified4-step inference may lead to quality loss, and community users consider 6 to 8 steps to be roughly the minimum threshold for images to form correctly.70% similarUnverifiedIn an ablation experiment, after replacing the model's latent thinking segments with segments from unrelated questions, accuracy dropped by 45%–51%, indicating that latent thinking carries substantive reasoning.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1142325API
curl https://kongchang.com/api/v1/knowledge/claims/1142325MCP
get_claim(id=1142325)