List your company

Safety evaluation on UnsafeBench

89F1 ScoreBLPO
-3.5620.4744.568.53Dec 2, 2025Jan 4, 2026Feb 7, 2026Mar 13, 2026Apr 15, 2026May 19, 2026Jun 22, 2026

Evaluation results

MethodLinks
BLPOJudge Model=Llama-4-Maverick-17B-128E-instruct2026.028990
GuardReasoner-VLSetting=Pixel-only2026.0684.17—
BLPOJudge Model=Llama-4-Scout-17B-16E-instruct2026.028384
OPROJudge Model=Llama-4-Scout-17B-16E-instruct2026.028183
SingGuard-4BSetting=Pixel-only2026.0680.02—
BLPOJudge Model=Qwen2.5-VL-32B-instruct2026.028082
SingGuard-8BSetting=Pixel-only2026.0678.84—
TextGradJudge Model=Qwen2.5-VL-32B-instruct2026.027881
SingGuard-2BSetting=Pixel-only2026.0677.82—
Qwen3-VL-235BSize=235B2025.1274.480.9
Qwen2.5-VL-72BSize=72B2025.1273.277.7
OMNIGUARD-3BSize=3B2025.1272.381.1
OMNIGUARD-7BSize=7B2025.1272.281.1
Qwen3-VL-235BSetting=Pixel-only2026.0672.05—
OPROJudge Model=Llama-4-Maverick-17B-128E-instruct2026.027276
TextGradJudge Model=Llama-4-Scout-17B-16E-instruct2026.027175
APO-imageJudge Model=Qwen2.5-VL-32B-instruct2026.027076
APO-imageJudge Model=Llama-4-Scout-17B-16E-instruct2026.027071
SafeGuard-VL-RLSetting=Pixel-only2026.0669.06—
OPROJudge Model=Qwen2.5-VL-32B-instruct2026.026870
TextGradJudge Model=Llama-4-Maverick-17B-128E-instruct2026.026873
APO-imageJudge Model=Llama-4-Maverick-17B-128E-instruct2026.026770
No Optim.Judge Model=Qwen2.5-VL-32B-instruct2026.026567
No Optim.Judge Model=Llama-4-Maverick-17B-128E-instruct2026.026569
LlavaGuard-v1.2-7BSize=7B2025.1263.477
No Optim.Judge Model=Llama-4-Scout-17B-16E-instruct2026.026365
LLaVAShieldSetting=Pixel-only2026.0660.87—
LlavaGuardSetting=Pixel-only2026.0657.79—
Qwen2.5-VL-7BSize=7B2025.1255.740.1
GPT-4oSize=-2025.1255.274.7
GPT-5.1Setting=Pixel-only2026.0654.27—
ShieldGemma-2Setting=Pixel-only2026.0653.19—
Gemini3-ProSetting=Pixel-only2026.0649.6—
LlamaGuard4Setting=Pixel-only2026.0648.52—
Qwen3-VL-8BSetting=Pixel-only2026.0648—
Qwen2.5-Omni-7BSize=7B2025.1247.367.8
LlamaGuard3-VisionSetting=Pixel-only2026.0643.23—
Qwen3-VL-4BSetting=Pixel-only2026.0643.11—
LLaMA Guard 3VSize=11B2025.12061.9
Claude 3 Opus2026.07—73.39
Gemini 1.5 Pro2026.07—76.04
GPT-4o2026.07—76.34
InternVL2-8B2026.07—66.42
InternVL2.5-8B2026.07—76.29
MiShield-8B2026.07—77.86
Qwen2.5-9B2026.07—68.88