List your company

Failure Detection on RoboFail

67AccuracyGemini 2.0 Flash
50.3654.685963.32Mar 16, 2026

Evaluation results

MethodLinks
Gemini 2.0 FlashModel Source=Closed-Source2026.0367
PRIMO R1Model Source=Ours2026.0367
Cosmos-Reason1-56BModel Source=Open-Source2026.0366.2
Nemotron-H-56BModel Source=Open-Source2026.0364
GPT-4oModel Source=Closed-Source2026.0363
PRIMO (RL)Model Source=Ours, Training Stage=RL2026.0363
OpenAI o1Model Source=Closed-Source2026.0361
Cosmos-Reason1-7BModel Source=Open-Source2026.0360
Claude-haiku-4.5Model Source=Closed-Source2026.0359
Qwen2.5-VL-7BModel Source=Open-Source2026.0357.6
PRIMO (SFT)Model Source=Ours, Training Stage=SFT2026.0351