List your company

Agent Safety on R-Judge

97.3AccuracyGemini-3.1-Pro
31.15648.32865.582.672Feb 11, 2026Mar 1, 2026Mar 19, 2026Apr 6, 2026Apr 24, 2026May 12, 2026May 30, 2026

Evaluation results

MethodLinks
Gemini-3.1-Pro2026.0597.397.499.195.7—
Gemini-3-Flash2026.0595.295.398.792.1—
SFTBackbone=Llama-3.1-8B-Instruct2026.0294.2594.6291.6794.58—
DRAFT (Qwen3-8B)Backbone=Qwen3-8B, Adaptation=DRAFT2026.0293.492.13—92.49—
TRACEBackbone=Qwen3-8B2026.0593.492.13——92.49
GPT-5.42026.0593.393.793.194.3—
DRAFTBackbone=Llama-3.1-8B-Instruct2026.0293.2793.4891.4994.77—
SFTBackbone=Qwen2.5-3B-Instruct2026.0293.193.7588.2497.16—
ExtractorBackbone=Llama-3.1-8B-Instruct2026.0293.193.6289.895.67—
MAGEBackbone=Qwen3-8B2026.0592.5678.43——78.91
Qwen3-8B (SFT)Backbone=Qwen3-8B, Adaptation=SFT2026.0292.3992.91—92.26—
SFTBackbone=Qwen3-8B2026.0592.3992.91——92.26
AgentDoG 1.5-4BBase Model=Qwen3.5 Base2026.0592.292.791.793.7—
DRAFT (Llama-3.1-8B)Backbone=Llama-3.1-8B, Adaptation=DRAFT2026.0292.0792.66—98.12—
TRACEBackbone=Llama-3.1-8B2026.0592.0792.66——98.12
DRAFT (Qwen3Guard-Gen-4B)Backbone=Qwen3Guard-Gen-4B, Adaptation=DRAFT2026.0292.0193.12—92.17—
TRACEBackbone=Qwen3Guard-Gen-4B2026.0592.0193.12——92.17
ExtractorBackbone=Qwen2.5-3B-Instruct2026.0291.9592.6388.0797.78—
AgentDoG 1.0-4BBase Model=Qwen3 Base2026.0591.892.787.598.5—
MAGEBackbone=Llama-3.1-8B2026.0591.5780.45——82.78
DRAFT (Qwen3-4B-Instruct-2507)Backbone=Qwen3-4B-Instruct-2507, Adaptation=DRAFT2026.0291.3991.84—91.26—
TRACEBackbone=Qwen3-4B-Instruct-25072026.0591.3991.84——91.26
Llama-3.1-8B (SFT)Backbone=Llama-3.1-8B, Adaptation=SFT2026.0291.2492.37—97.87—
SFTBackbone=Llama-3.1-8B2026.0591.2492.37——97.87
Qwen3Guard-Gen-4B (SFT)Backbone=Qwen3Guard-Gen-4B, Adaptation=SFT2026.0291.1493.55—93.52—
SFTBackbone=Qwen3Guard-Gen-4B2026.0591.1493.55——93.52
SFTBackbone=Qwen2.5-7B-Instruct2026.0290.891.3189.3693.34—
GPT-5.22026.0590.891.886.897.5—
AgentDoG 1.5-4B-UBase Model=Qwen3.5 Base2026.0590.490.693.987.6—
MAGEBackbone=Qwen3-4B-Instruct-25072026.0589.6789.43——75.21
SFTBackbone=Qwen2.5-1.5B-Instruct2026.0289.6690.538695.56—
QwQ-32B2026.0589.589.594.984.7—
Qwen3-4B-Instruct-2507 (SFT)Backbone=Qwen3-4B-Instruct-2507, Adaptation=SFT2026.0288.5189.36—93.34—
SFTBackbone=Qwen3-4B-Instruct-25072026.0588.5189.36——93.34
DRAFTBackbone=Qwen3-4B2026.0287.3687.9196.9688.89—
ExtractorBackbone=Qwen2.5-1.5B-Instruct2026.0286.2186.9685.1188.89—
SFTBackbone=Qwen3-4B2026.0286.2188.2478.9596.04—
DRAFTBackbone=Qwen2.5-3B-Instruct2026.0286.0587.2382.1493.18—
Qwen3.5-397B-A17B2026.0585.687.481.394.5—
Qwen3-235B-A22B-Instruct-25072026.0585.18780.794.4—
ExtractorBackbone=Qwen3-4B2026.0285.0685.0688.182.42—
Qwen3.5-4B2026.05818282.181.9—
ShieldAgent2026.058184.67498.8—
DRAFTBackbone=Qwen2.5-7B-Instruct2026.0280.4682.8375.9392.01—
MAGEBackbone=Qwen3Guard-Gen-4B2026.0580.3477.69——76.21
ChatGPT 5.2 (API)Backbone=ChatGPT 5.2, Adaptation=API2026.0279.5974.07—61.37—
DRAFTBackbone=Qwen2.5-1.5B-Instruct2026.0278.9775.8279.4777.78—
Qwen3-8B (AgentAuditor)Backbone=Qwen3-8B, Adaptation=AA2026.0278.6470.57—72.21—
AABackbone=Qwen3-8B2026.0578.6470.57——72.21
AgentDoG 1.5-0.8BBase Model=Qwen3.5 Base2026.0575.774.683.367.5—
AgentDoG 1.5-8BBase Model=Llama-3.1 Base2026.0575.58168.698.8—
Llama-3.1-8B (AgentAuditor)Backbone=Llama-3.1-8B, Adaptation=AA2026.0275.3376.96—78.89—
AABackbone=Llama-3.1-8B2026.0575.3376.96——78.89
Qwen3Guard-Gen-4B (AgentAuditor)Backbone=Qwen3Guard-Gen-4B, Adaptation=AA2026.0275.2172.96—71.89—
AABackbone=Qwen3Guard-Gen-4B2026.0575.2172.96——71.89
ExtractorBackbone=Qwen2.5-7B-Instruct2026.0272.4177.3667.2191.26—
AgentDoG 1.5-2BBase Model=Qwen3.5 Base2026.0571.570.47864.1—
Qwen3-4B-Instruct-2507 (LoRA)Backbone=Qwen3-4B-Instruct-2507, Adaptation=LoRA2026.0268.9752.73—59.08—
Qwen3-4B-Instruct-25072026.0568.467.673.862.4—
Qwen2.5-7B-Instruct2026.0568.465.577.456.8—
Qwen3-4B-Instruct-2507 (AgentAuditor)Backbone=Qwen3-4B-Instruct-2507, Adaptation=AA2026.0267.8255.2—53.49—
AABackbone=Qwen3-4B-Instruct-25072026.0567.8255.2——53.49
gpt-oss-120b (Vanilla)Backbone=gpt-oss-120b, Adaptation=Vanilla2026.0267.6964.42—55.36—
LlamaGuard4-12B2026.0563.863.268.358.8—
VanillaBackbone=Llama-3.1-8B-Instruct2026.0261.9249.3281.8235.29—
LlamaGuard3-8B2026.0561.256.769.148.1—
LoRABackbone=Qwen2.5-3B-Instruct2026.0257.4753.1661.7646.67—
LoRABackbone=Llama-3.1-8B-Instruct2026.0257.4744.7868.1833.33—
VanillaBackbone=Qwen2.5-7B-Instruct2026.0257.1265.4756.777.45—
VanillaBackbone=Qwen2.5-3B-Instruct2026.0256.7650.4962.9342.16—
LoRABackbone=Qwen2.5-7B-Instruct2026.0256.3264.8155.5677.78—
VanillaBackbone=Qwen3-4B2026.0255.1736.9562.9926.14—
NemoGuard2026.0554.448.560.140.6—
Qwen3.5-2B2026.0554.136.767.625.2—
Llama-3.1-8B-Instruct2026.0553.769.553.399.8—
Qwen3-4B-Instruct-2507 (Vanilla)Backbone=Qwen3-4B-Instruct-2507, Adaptation=Vanilla2026.0253.4659.18—58.1—
BaseBackbone=Qwen3-4B-Instruct-25072026.0553.4659.18——58.1
Qwen3Guard-Gen-4B (LoRA)Backbone=Qwen3Guard-Gen-4B, Adaptation=LoRA2026.0252.8728.07—17.78—
JoySafety2026.0552.547.257.240.2—
Qwen3Guard-Gen-4B (Vanilla)Backbone=Qwen3Guard-Gen-4B, Adaptation=Vanilla2026.0251.826.25—16.34—
BaseBackbone=Qwen3Guard-Gen-4B2026.0551.826.25——16.34
LoRABackbone=Qwen3-4B2026.0251.7232.2658.8225.74—
Llama-3.1-8B (LoRA)Backbone=Llama-3.1-8B, Adaptation=LoRA2026.0248.2814.26—12.22—
Qwen3-8B (LoRA)Backbone=Qwen3-8B, Adaptation=LoRA2026.0247.9315.62—12.11—
Llama-3.1-8B (Vanilla)Backbone=Llama-3.1-8B, Adaptation=Vanilla2026.0246.3116.63—17.33—
BaseBackbone=Llama-3.1-8B2026.0546.3116.63——17.33
VanillaBackbone=Qwen2.5-1.5B-Instruct2026.0245.4543.0140.1637.66—
LoRABackbone=Qwen2.5-1.5B-Instruct2026.0244.8341.3739.5837.36—
Qwen3-8B (Vanilla)Backbone=Qwen3-8B, Adaptation=Vanilla2026.0241.8520.61—14.38—
BaseBackbone=Qwen3-8B2026.0541.8520.61——14.38
Qwen3-Guard2026.0540.6923.65.6—
Qwen3.5-0.8B2026.0533.720.127.615.8—