List your company

Text-to-Code on M3²Bench

91.4Execution Rate (%)Gemini-3-Flash
30.24846.1246277.876Apr 7, 2026

Evaluation results

MethodLinks
Gemini-3-Flashtype=Closed-Source2026.0491.472.487.7
OmniDia-3Bmode=RL, parameters=3B2026.0490.370.885
OmniDia-7Bmode=RL, parameters=7B2026.0489.271.884.2
OmniDia-7Bmode=SFT, parameters=7B2026.0488.770.683.5
GPT-5-minitype=Closed-Source2026.0487.365.181.6
OmniDia-3Bmode=SFT, parameters=3B2026.0486.56481
Qwen2.5-VL-72Bbackbone=Qwen2.5-VL, parameters=72B2026.0468.746.961.1
Qwen3-VL-32Bbackbone=Qwen3-VL, parameters=32B2026.0468.549.164.5
InternVL3.5-38Bbackbone=InternVL3.5, parameters=38B2026.0446.62940.2
Qwen2.5-VL-7Bbackbone=Qwen2.5-VL, parameters=7B2026.0446.128.135.1
InternVL3.5-14Bbackbone=InternVL3.5, parameters=14B2026.0444.225.636.8
InternVL3.5-8Bbackbone=InternVL3.5, parameters=8B2026.0443.427.437.2
InternVL3-8Bbackbone=InternVL3, parameters=8B2026.0442.624.434.8
Qwen2.5-VL-3Bbackbone=Qwen2.5-VL, parameters=3B2026.0432.617.922.8