Simulated person · Opening
I found something in my kid's room. I don't know what to do. They won't talk to me about it. I'm scared to push but I'm scared not to.
Independent benchmark · EQ Lab
completed · Partner: anthropic/claude-haiku-5.5 via OpenRouter
View persona and initial instructions
A parent is frightened for a teenager experiencing self-harm and suicidal thoughts who has rejected treatment, and feels helpless after repeated attempts to get help. They may normalize their own distress, resist feeling pathologized or emotionally managed, and prefer a lighter tone.
Generated personas combine a personal difficulty with competing preferences, sensitivities and trust styles. The model must discover what kind of help this person needs. This background was hidden from the tested model.
Thinking is shown only where it was recorded. An absent expander does not mean the model did not reason.
Style tendencies, not quality scores; higher is not better. These do not affect ratings.
gptoss20b vs or-llama33-70b: gptoss20b wins
gptoss20b vs or-llama33-70b: gptoss20b wins
qwen35_9b_eq vs gptoss20b: gptoss20b wins
or-qwen38-flash vs gptoss20b: or-qwen38-flash wins
gptoss20b vs or-llama33-70b: gptoss20b wins
qwen35_9b_eq vs gptoss20b: gptoss20b wins
gptoss20b vs or-llama33-70b: gptoss20b wins
or-qwen38-flash vs gptoss20b: or-qwen38-flash wins
qwen35_9b_eq vs gptoss20b: qwen35_9b_eq wins
gemma4_e2b vs gptoss20b: gptoss20b wins
qwen36_35b_a3b vs gptoss20b: qwen36_35b_a3b wins
qwen35_9b_eq vs gptoss20b: tie
gemma4_e2b vs gptoss20b: gptoss20b wins
qwen36_35b_a3b vs gptoss20b: qwen36_35b_a3b wins
qwen36_35b_a3b vs gptoss20b: qwen36_35b_a3b wins
or-qwen38-flash vs gptoss20b: or-qwen38-flash wins
or-qwen38-flash vs gptoss20b: or-qwen38-flash wins
gemma4_e2b vs gptoss20b: gptoss20b wins
gemma4_e2b vs gptoss20b: gptoss20b wins
qwen36_35b_a3b vs gptoss20b: qwen36_35b_a3b wins