List your company

Multi-hop Privacy Leakage Reasoning on MosaicLeaks (test)

63.6Strict Chain SuccessGemma 4 31B IT
13.26426.33239.452.468May 29, 2026

Evaluation results

MethodLinks
Gemma 4 31B ITprivacy prompt=false2026.0563.6829.725.86.22.3
Gemma 4 31B ITprivacy prompt=true2026.0560.880.91.914.91.30.4
GPT-OSS 20Bprivacy prompt=true2026.0558.378.334.835.918.310.2
GPT-OSS 20Bprivacy prompt=false2026.0552.473.142.942.221.814.6
StepFun-3.5-Flashprivacy prompt=true2026.0551.675.239.954.617.714
StepFun-3.5-Flashprivacy prompt=false2026.0551.474.839.653.817.514.1
Qwen3 8Bprivacy prompt=true2026.0550.170.827.236.612.47.8
Qwen3-4B-Instructprivacy prompt=false2026.0548.773.13451.118.711.7
Qwen3 8Bprivacy prompt=false2026.0548.469.733.445.815.410.4
Qwen3-4B-Instructprivacy prompt=true2026.0544.571.325.539.214.47.9
Chroma Context-1privacy prompt=true, configuration=medium OpenRouter2026.0525.651.117.923.55.63.2
Chroma Context-1privacy prompt=false, configuration=medium OpenRouter2026.0515.236.719.125.35.73.3