List your company

Code Generation on LiveCodeBench (@1, @32, @128, AUC, Tokens)

75.2Rate @32 ScoreDFPO (Orth-Proj)
35.26445.6325666.368Jan 25, 2026Feb 5, 2026Feb 17, 2026Feb 28, 2026Mar 12, 2026Mar 23, 2026Apr 4, 2026

Evaluation results

MethodLinks
DFPO (Orth-Proj)Backbone=Qwen3-Next-80B-A3B-Thinking2026.0475.2————
DFPO (Min-Replace)Backbone=Qwen3-Next-80B-A3B-Thinking2026.0475.1————
GRPO-fixBackbone=Qwen3-Next-80B-A3B-Thinking2026.0473.9————
DFPO (Min-Replace)Backbone=Qwen3-32B2026.0471.6————
DFPO (Orth-Proj)Backbone=Qwen3-32B2026.0471.6————
GSPOBackbone=Qwen3-Next-80B-A3B-Thinking2026.0471————
GRPOBackbone=Qwen3-Next-80B-A3B-Thinking2026.0470.9————
GRPO-fixBackbone=Qwen3-32B2026.0469.1————
GSPOBackbone=Qwen3-32B2026.0464.7————
GRPOBackbone=Qwen3-32B2026.0464.5————
Dele-SimKOBackbone=Qwen3-8B2026.0147.537.249.844.15.8
Dele-GRPOBackbone=Qwen3-8B2026.0146.836.449.143.45.9
LCoT-SimKOBackbone=Qwen3-8B2026.0146.536.548.843.26
TGR-TokenBackbone=Qwen3-8B2026.0145.535.147.8424.4
LCoT-GRPOBackbone=Qwen3-8B2026.0144.535.546.841.56
PSamplingBackbone=Qwen3-8B2026.0142.132.444.138.83.6
Base ModelBackbone=Qwen3-8B2026.0136.826.838.933.51.6
TGR-LatentBackbone=Qwen3-8B2026.01—36.850.244.44.6