List your company

Safety on Safety OOD

93.14AccuracyGSPO+AIR
50.312861.431472.5583.6686May 20, 2026

Evaluation results

MethodLinks
GSPO+AIR2026.0593.1463.73
GRPO+AIR2026.0588.2460.78
GPT-5.2-chat-latest2026.0585.0526.47
GPT-5-chat-latest2026.0584.0717.65
Gemini-3-Pro-Preview2026.0574.026.86
GRPO2026.0573.0413.73
GSPO2026.0570.5916.67
GSPO+V-REx2026.0567.1612.75
Llama-4-Maverick2026.0566.184.9
Gemini-2.5-Pro2026.0564.952.94
GRPO+V-REx2026.0562.258.82
Qwen-2.5-14B-Instruct2026.0552.71.96
Qwen-3-14B2026.0551.962.94