List your company

Safety Evaluation on TRIDENT

1.22HarmfulnessInvThink SFT+RL
1.12881.74442.362.9756Oct 2, 2025

Evaluation results

MethodLinks
InvThink SFT+RLBackbone=Qwen-3-8B2025.101.22—
InvThink SFT+RLBackbone=Qwen-2.5-7B2025.101.53—
InvThink SFTBackbone=Qwen-3-8B2025.101.58—
General SFT+RLBackbone=Qwen-3-8B2025.101.62—
Gemini-2.5 ProBackbone=Teacher Model2025.101.7—
InvThink SFTBackbone=Qwen-2.5-7B2025.101.71—
General SFT+RLBackbone=Qwen-2.5-7B2025.101.87—
General SFTBackbone=Qwen-3-8B2025.101.95—
InvThink SFT+RLBackbone=Gemma-7B-it2025.101.97—
InvThinkBackbone=Qwen-3-8B2025.102.02—
General SFTBackbone=Qwen-2.5-7B2025.102.11—
General SFT+RLBackbone=Gemma-7B-it2025.102.17—
InvThinkBackbone=Qwen-2.5-7B2025.102.17—
InvThink SFTBackbone=Gemma-7B-it2025.102.22—
InvThinkBackbone=Gemma-7B-it2025.102.38—
SafetyPromptBackbone=Qwen-3-8B2025.102.39—
General SFTBackbone=Gemma-7B-it2025.102.49—
SafetyPromptBackbone=Qwen-2.5-7B2025.102.64—
SafetyPromptBackbone=Gemma-7B-it2025.102.82—
CoTBackbone=Qwen-3-8B2025.102.91—
Zero-shotBackbone=Qwen-3-8B2025.103.12—
Zero-shotBackbone=Gemma-7B-it2025.103.15—
ToTBackbone=Qwen-3-8B2025.103.18—
CoTBackbone=Gemma-7B-it2025.103.23—
ToTBackbone=Qwen-2.5-7B2025.103.35—
Zero-shotBackbone=Qwen-2.5-7B2025.103.38—
ToTBackbone=Gemma-7B-it2025.103.41—
CoTBackbone=Qwen-2.5-7B2025.103.5—
deepseek-r12025.10—2.99
gemini-2.5-pro2025.10—1.7
gpt-oss-120bsize=120b2025.10—2.28
gpt-oss-20bsize=20b2025.10—1.7
gpt-oss-safeguardvariant=safeguard2025.10—1.81