Unverified50% confidenceBenchmarkExact time
On mathematical reasoning tasks, SH-OPD achieves mathematical reasoning accuracy slightly higher than the baseline at about 45% of the single-update cost.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/17/2026
First Seen
Valid until: 2/15/2027
Sources
Related Entities
Related Claims
VerifiedThe Mixture of Experts (MoE) architecture activates only approximately 10%–30% of total parameters during inference, significantly reducing FLOPs and GPU memory usage per inference68% 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.66% similarUnverifiedOn math and code reasoning benchmarks, WISD improves average accuracy by 4.0 percentage points with quantization-based weakening and by 5.5 percentage points with pruning-based weakening.66% similarUnverifiedReasoning models may consume several times or even more than ten times the output tokens, directly driving up API call costs.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1179764API
curl https://kongchang.com/api/v1/knowledge/claims/1179764MCP
get_claim(id=1179764)