List your company

Code Generation on MBPP+ (Accuracy & Improvement)

76.9AccuracyPrimitives-based MAS
38.21248.25658.368.344Feb 3, 2026Feb 10, 2026Feb 18, 2026Feb 25, 2026Mar 5, 2026Mar 12, 2026Mar 20, 2026

Evaluation results

MethodLinks
Primitives-based MASBackbone=Qwen3-14B2026.0276.98.4
Primitives-based MASBackbone=Qwen3-4B2026.0276.412.9
LatentMASBackbone=Qwen3-14B2026.0275.77.2
VotingBackbone=Qwen3-14B2026.0275.77.2
PlanningBackbone=Qwen3-14B2026.0274.66.1
VotingBackbone=Qwen3-4B2026.0273.810.3
ReviewBackbone=Qwen3-14B2026.0273.85.3
LatentMASBackbone=Qwen3-4B2026.0273.510
TextMASBackbone=Qwen3-14B2026.0272.84.3
ReviewBackbone=Qwen3-4B2026.0272.28.7
PlanningBackbone=Qwen3-4B2026.0272.28.7
TextMASBackbone=Qwen3-4B2026.0269.86.3
SingleBackbone=Qwen3-14B2026.0268.5—
Qwen3Parameters=8B2026.0367.9—
Qwen2.5-7BBase Model=Qwen2.5-7B2026.0367.5—
Qwen3-8BBase Model=Qwen3-8B2026.0366.4—
SDARBase Model=Qwen3-8B2026.0366.40
TraDOBase Model=Qwen3-8B2026.0366.40
Gemma3Parameters=12B2026.0366.1—
Qwen2.5Parameters=4B2026.0365.9—
LoopRPTParameters=2.6B2026.0365.08—
OuroParameters=2.6B2026.0364.81—
SingleBackbone=Qwen3-4B2026.0263.5—
Qwen2.5Parameters=7B2026.0363.5—
Gemma3Parameters=3B2026.0354.2—
dTRPOBase Model=Qwen2.5-7B2026.0351.615.9
Llama3.1Parameters=8B2026.0351.6—
Qwen3Parameters=4B2026.0351.1—
Llama3.2Parameters=3B2026.0339.7—