List your company

Multimodal tool-use on GTA

67.66Answer AccuracyReGRPO (default, λval = 0)
11.957626.418840.8855.3412Apr 30, 2025Jul 10, 2025Sep 19, 2025Nov 29, 2025Feb 8, 2026Apr 20, 2026Jun 30, 2026

Evaluation results

MethodLinks
ReGRPO (default, λval = 0)Controller=MAT-Qwen2-VL-7B2026.0667.6676.3593.77
GPT-5Params=-2026.0360.968.398.7
SPORT AgentController=Tuned-Qwen2-VL-7B2025.0460.2672.4191.87
SPORT AgentController=Tuned-Qwen2-VL-7B2026.0660.2672.4191.87
GPT-4.1Params=-2026.0358.465.194.3
HF AgentController=GPT-4o-mini2025.0457.6956.1100
HF AgentController=GPT-4o-mini2026.0657.6956.1100
GPT-4oParams=-2026.0357.163.495.1
HF AgentController=GPT-4o2025.0457.0563.4195.12
HF AgentController=GPT-4o2026.0657.0563.4195.12
TraceR1Params=8B2026.0356.765.787.4
T3-AgentController=MAT-Qwen2-VL-7B2025.0453.8564.6384.32
T3-AgentController=MAT-Qwen2-VL-7B2026.0653.8564.6384.32
T3-AgentParams=7B2026.0353.864.684.3
T3-AgentController=MAT-MiniCPM-V-8.5B2025.0452.5665.8580.49
T3-AgentController=MAT-MiniCPM-V-8.5B2026.0652.5665.8580.49
Qwen3-VL-8BParams=8B2026.0349.256.874.2
Qwen2.5-VL-14BParams=14B2026.0346.855.469.8
Lego AgentController=GPT-42025.0446.59——
Lego AgentController=GPT-42026.0646.59——
Qwen2.5-VL-7BParams=7B2026.0344.250.669.1
HF AgentController=Qwen2-VL-7B2025.0442.3144.8565.19
HF AgentController=Qwen2-VL-7B2026.0642.3144.8565.19
Lego AgentController=GPT-4o2025.0441.52——
Lego AgentController=GPT-4o2026.0641.52——
HF AgentController=MiniCPM-V-8.5B2025.0433.9736.5956.1
HF AgentController=MiniCPM-V-8.5B2026.0633.9736.5956.1
HF AgentController=InternVL2-8B2025.0432.0536.7552.18
HF AgentController=InternVL2-8B2026.0632.0536.7552.18
DeepSeek-VL2Params=72B2026.0323.249.457.2
LLAVA-NeXT-8BParams=8B2026.0314.114.925.1
HF AgentController=LLaVA-NeXT-8B2025.0414.114.9725.08
HF AgentController=LLaVA-NeXT-8B2026.0614.114.9725.08