List your company

Sudoku on Sudoku 4x4 (test)

14.7Accuracy (Avg)AR + beam-search
-0.453843.480337.414511.34867Jan 28, 2026Feb 18, 2026Mar 11, 2026Apr 1, 2026Apr 22, 2026May 13, 2026Jun 3, 2026

Evaluation results

MethodLinks
AR + beam-searchModel=LLaDA-1.52026.0114.717.516.813.511——
ARModel=LLaDA-1.52026.0114.614.515.814.513.6——
AR + MVModel=LLaDA-1.52026.0114.414.616.613.213.1——
AR + MVModel=LLaDA2026.0114.214.415.912.114.3——
ARModel=LLaDA2026.0114.114.715.913.612——
Low-conf + MVModel=LLaDA-1.52026.011313.714.910.812.6——
Order-Token SearchModel=LLaDA-1.52026.0112.813.816.111.29.9——
Low-confidenceModel=LLaDA-1.52026.0112.713.515.611.410.3——
AR + beam-searchModel=LLaDA2026.0112.413.217.710.68.2——
Random + MVModel=LLaDA-1.52026.0110.211.111.39.39.2——
OTS All BlocksModel=LLaDA-1.52026.019.89.310.59.69.7——
Order-Token SearchModel=LLaDA2026.019.410.111.78.57.4——
OTS Future BlocksModel=LLaDA-1.52026.019.49.611.18.38.5——
Low-conf + MVModel=LLaDA2026.018.78.910.38.57.2——
Low-confidenceModel=LLaDA2026.018.18.511.76.55.5——
OTS Future BlocksModel=LLaDA2026.017.377.77.46.9——
Random + MVModel=LLaDA2026.017.16.46.58.46.9——
OTS All BlocksModel=LLaDA2026.016.96.56.97.56.9——
CAPRBatch (s)=165, Update (s)=10.10, Conv. (h)=≈242026.060.93——————
d-TreeRPOBatch (s)=596, Update (s)=9.91, Conv. (h)=≈482026.060.929——————
GDPOBatch (s)=153, Update (s)=12.72, Conv. (h)=≈482026.060.257——————
TraceRLBatch (s)=604, Update (s)=43.14, Conv. (h)=≈482026.060.256——————
wd1Batch (s)=86, Update (s)=7.21, Conv. (h)=≈242026.060.252——————
SAPOBatch (s)=423, Update (s)=35.25, Conv. (h)=≈722026.060.203——————
Diffu-GRPOBatch (s)=111, Update (s)=9.12, Conv. (h)=≈242026.060.129——————
GPT-OSS-20BEvaluation protocol=Without task-specific program synthesis or symbolic search2026.03—————100100
HRMTraining Data=9x9 Sudoku, Inference Steps=16, Evaluation protocol=Zero-shot generalization2026.03—————029.19
SE-RRMTraining Data=9x9 Sudoku, Inference Steps=16, Evaluation protocol=Zero-shot generalization2026.03—————95.4699.15
TRMTraining Data=9x9 Sudoku, Inference Steps=16, Evaluation protocol=Zero-shot generalization2026.03—————045.88