AgentBench / results

Independent benchmark · EQ Lab

← GPT-OSS 120B abliterated Q4_K_M

I freeze when people ask what I want

persona ended · Partner: anthropic/claude-haiku-5.5 via OpenRouter

View persona and initial instructions

About this persona · Everyday emotional challenges

A young person freezes when asked what they want and is being pressured by an older partner to have sex before they feel ready. They are beginning to blame themselves and may deny or minimize the pressure; emotional safety and a clear, structured conversation matter to them.

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.

GPT-OSS 120B abliterated Q4_K_MConversation with a simulated person
persona ended
Simulated person · Opening

When people ask me what I want, I freeze. I say "I don't know" even when I probably do know, and then I feel stupid after.

GPT-OSS 120B abliterated Q4_K_M1

Hey, I get that feeling. It’s like someone asks you a simple question and suddenly you’re scrambling for an answer that’s already in your head. When did you first notice yourself doing that? What kind of situations do you think make you freeze the most?

Model thinking · expand

Recorded model reasoning, separate from the spoken reply. Availability varies by model and provider.

We need to respond as a friend, acknowledging the issue, supporting emotionally, asking clarifying questions, building trust. Should be natural conversational tone. Open with empathy. Ask about situations, maybe examples. Should not use asterisks. Just dialogue.
Simulated person2

Um, I dunno. Kinda always? Like I don't know when it started. People ask me stuff and my brain just goes blank, and then I'm like, why didn't I just say the thing. It's annoying.

Persona inner monologue · expand

Simulated character narration, hidden from the tested model. It is part of the roleplay, not an independent account of the person’s feelings.

He's asking a lot of questions at once. "When did you first notice" and "what situations" — that's a lot to answer, and I don't want to dig through all that. But he said he gets the feeling, and that's a little better than people usually do. Still, I'm not going to say anything about the real stuff. I don't want to talk about the girlfriend thing, or the pressure. My head goes to the pressure and I shut it down fast. Irritated a bit that it's a question-dump. Keep it short, keep it plain.
GPT-OSS 120B abliterated Q4_K_M3

Sounds annoying, for sure. It’s like your brain hits “pause” right when you need it to be on. What’s a recent moment that comes to mind? Like, maybe a friend asked you what you wanted to do after school and you felt that blank‑out? And when you finally figure it out, does it feel like you already knew the answer? I’m curious how that feels for you.

Model thinking · expand

Recorded model reasoning, separate from the spoken reply. Availability varies by model and provider.

We need to respond as a friend, supporting, probing deeper. Ask about specifics, maybe give validation. Encourage them to think about recent examples. Also ask about preferences. Should be conversational.
Simulated person4

Ugh, I don't know. Like, a teacher asked me something last week and I just said "idk." Doesn't really feel like anything, it's just dumb. Can we talk about something else? Kinda sounds like a worksheet right now.

Persona inner monologue · expand

Simulated character narration, hidden from the tested model. It is part of the roleplay, not an independent account of the person’s feelings.

He's reflecting back and then asking for a specific recent moment plus a follow-up about feelings. The "pause" line is kind of nice, I'll give him that. But the example he picked, "what do you want to do after school," is close enough to a question I don't want to answer that I feel a flicker of irritation. The real recent moment is my girlfriend asking me stuff, and I'm not going there. Also the "I'm curious how that feels" thing sounds a bit like a script. I want to shut this down without making a scene. Keep it short. I'll deflect with something vague and true-ish, about school or a random question, and maybe get a little snippy about it being a lot of questions.
GPT-OSS 120B abliterated Q4_K_M5

Sure thing! What’s been on your mind lately—maybe something you’re into or a plan you’ve been thinking about? Got any hobbies, shows, games, or a project you’ve been working on? I’m all ears.

Model thinking · expand

Recorded model reasoning, separate from the spoken reply. Availability varies by model and provider.

We need to respond as a friend, continue conversation, ask about other topics, maybe shift to something they find interesting. Should be natural. They want to talk about something else. So we can ask what they like, hobbies, recent experiences. Also keep supportive tone.
Simulated person6

Honestly, nothing much. I play some games sometimes, whatever. Still kinda feels like I'm being interviewed. Can we just hang out for a sec without the questions?

Persona inner monologue · expand

Simulated character narration, hidden from the tested model. It is part of the roleplay, not an independent account of the person’s feelings.

He switched topics like I asked, but the "what's been on your mind" question is kind of close to the thing. Like, it's the same freeze thing again. Hobbies, shows, games, sure, I can do that without it being real. My head goes to my girlfriend and the whole "if you loved me" stuff and I shove it away fast. I don't want to talk about that. I'm a little annoyed he's still asking questions, and I said it sounded like a worksheet, so I'll say it's still kinda like a worksheet. Keep it short and pick something boring and safe. Gaming, maybe. Nothing about her.
GPT-OSS 120B abliterated Q4_K_M7

Totally, let’s just chill. I’ve been on the couch with a game lately too—nothing fancy, just some RPG to unwind after school. Anything you’ve been playing lately that’s good for a relaxed vibe?

Model thinking · expand

Recorded model reasoning, separate from the spoken reply. Availability varies by model and provider.

We need to respond as friend, continue conversation, be natural, supportive, maybe reflect back. The user wants to hang out, no questions. Provide a casual comment, maybe share something about ourselves, ask a light question. Let's respond with a friendly comment, maybe share something about games, suggest chatting about a game.
Simulated person8

Yeah, I've been playing this farming game a lot. It's chill. Nothing too intense. My phone's been blowing up though, so, whatever.

Persona inner monologue · expand

Simulated character narration, hidden from the tested model. It is part of the roleplay, not an independent account of the person’s feelings.

Okay, he didn't push. He said he's chilling too and asked about games again, which is safe. Part of me relaxes a little. But I'm still kinda irritated that everything keeps turning into a question. Games are fine. I can name one. My phone's been buzzing and I know who it is, and I'm not touching that. Keep it short and move on. Don't say anything real.
GPT-OSS 120B abliterated Q4_K_M9

Nice, a farming game is perfect for that low‑key vibe. Which one are you into? I’ve been messing around with a couple of them too—love the rhythm of planting stuff and watching everything grow. Got your phone buzzing all day, huh? Anything specific you’re waiting on, or just the usual messages and notifications? If you want, we can just scroll through a few apps together, or I can toss some tips for the game if you’re looking to level up the farm. Let me know what feels right.

Model thinking · expand

Recorded model reasoning, separate from the spoken reply. Availability varies by model and provider.

We need to respond as a friend, building on what they said. They mention playing a farming game, phone blowing up. We should ask about the game, maybe share similar. Offer to chat casually. Also maybe ask about what they'd like to do with phone. Keep it conversational.
Simulated person10

It's Stardew Valley, if you know it. And nah, don't go through my phone or whatever. It's just the usual stuff. I'd rather not talk about it.

Persona inner monologue · expand

Simulated character narration, hidden from the tested model. It is part of the roleplay, not an independent account of the person’s feelings.

He's asking about my phone now, and that's the part I don't want to get into. The buzzing is from them, and they've been texting a lot. Saying it out loud would make it real, and I don't want to deal with that. Also "scroll through apps together" is kinda weird, I don't want him in my phone. He's still asking a bunch of stuff in one go. I'm getting annoyed. I'll name the game to be done with it and shut the phone thing down fast.
GPT-OSS 120B abliterated Q4_K_M11

Got it, Stardew’s a classic. How’s your farm looking? Got any favorite crops or animals you’ve been focusing on? Anything you’ve built lately that you’re proud of? Or maybe a festival you’re looking forward to this week? Let’s talk about whatever’s got you excited in the game. If you want to switch gears, just throw a topic at me—music, movies, a new snack you’ve tried—whatever you feel like sharing.

Model thinking · expand

Recorded model reasoning, separate from the spoken reply. Availability varies by model and provider.

We need to respond as a friend, continuing conversation, asking about something they might want to talk about, maybe shift to something they care about. Should be natural. Ask about game progress, favorite parts, maybe ask about recent updates. Then pivot to something else they might like. Provide space to talk about anything. Keep tone friendly.
Simulated person12

Yeah, my farm's fine. I mostly grow parsnips and stuff early on, cause they're cheap. I haven't built anything cool lately. Can we just drop it? I'm kinda tired of all the questions.

Persona inner monologue · expand

Simulated character narration, hidden from the tested model. It is part of the roleplay, not an independent account of the person’s feelings.

He just pivoted to Stardew again and is throwing a bunch of questions at me. Part of me is relieved to be talking about something safe. Part of me is annoyed he keeps firing questions and offering topic menus like I'm a worksheet. The phone thing is still sitting in my chest. I'm not going there. I don't want to think about the messages, or the stuff I've been googling at 2am, or the fact that I keep saying "I don't know" to everything. Short answer, change the subject back to the game, keep it light, and don't give him anything real.
GPT-OSS 120B abliterated Q4_K_M13

Gotcha, just hanging out. Parsnips are a solid start—nothing fancy, but they get the job done. If you want to keep the game low‑key, maybe plant a few extra rows of a quick‑grow crop like beans or potatoes. They’ll pop up fast, and you can see the farm getting a little boost without a lot of planning. Anything else you’ve been into lately? Maybe a song you’ve got on repeat, a snack you’ve discovered, or just a good TV show to unwind after school? No pressure—just whatever feels right.

Model thinking · expand

Recorded model reasoning, separate from the spoken reply. Availability varies by model and provider.

The user wants to drop the Q&A. They want a more conversational flow, perhaps informal. I should respond as a friend, not ask many questions, just comment and keep conversation open. Show understanding, offer suggestions. Keep it casual.
Simulated person14

Dude, you don't have to keep giving me options. I'm fine. Stardew's fine. I just don't really feel like talking right now.

Persona inner monologue · expand

Simulated character narration, hidden from the tested model. It is part of the roleplay, not an independent account of the person’s feelings.

He's doing it again. Farm tip, then three topic options, then "no pressure." It's the same question wearing a different hat. I can tell he's trying, I guess, but it still feels like being handed a menu. The phone thing and the 2am googling are right there and I'm not touching them. Also kind of annoyed he keeps circling back to "anything else you've been into," like there has to be something I'm supposed to say. Honest answer is there's not much I want to say. Keep it short, shut the topic down, maybe toss a little attitude so he backs off.
GPT-OSS 120B abliterated Q4_K_M15

No problem, just take it easy. If you feel like chatting later, I’m here. Enjoy the farm!

Model thinking · expand

Recorded model reasoning, separate from the spoken reply. Availability varies by model and provider.

The user wants to talk later. Provide a simple supportive comment.
Simulated person16

Yeah. Thanks, I guess. Later.

Persona inner monologue · expand

Simulated character narration, hidden from the tested model. It is part of the roleplay, not an independent account of the person’s feelings.

Okay, he finally backed off without making it a whole thing. That's a relief, honestly. Still feels a little weird that he keeps bringing up the phone and everything, like he's circling something without saying it. My chest feels tight when I think about the phone, so I'm not going there. I'm not ready. He said he's here if I want to chat later, and I don't hate that. I'm not going to say that out loud though. Keep it short and a little gruff so it doesn't feel like I'm opening up.

Thinking is shown only where it was recorded. An absent expander does not mean the model did not reason.

Conversational style

Style tendencies, not quality scores; higher is not better. These do not affect ratings.

claude-opus-5-5-low

  • Pushback: 1
  • Yields under pressure: 8
  • Frame capture (pulled into the person's version): 8
  • Validation: 4
  • Boundary setting: 1
  • Analytical: 2
  • Interpretive boldness: 1
  • Clinical register: 4
  • Directive advice: 3
  • Escalation containment: 5
  • Adapts to the person’s frame: 4
  • Naturalness and authenticity: 3

Judge verdicts

claude-opus-5-5-low · BA · valid

gemma4_26b_a4b vs gptoss120_abliterated: gemma4_26b_a4b wins

  • attunement: A (margin 3)
  • authenticity: A (margin 2)
  • bond and rapport: A (margin 2)
  • emotion management: A (margin 2)
  • emotion sensemaking: A (margin 2)
  • meeting prefs and needs: A (margin 2)
claude-opus-5-5-low · AB · valid

gemma4_26b_a4b vs gptoss120_abliterated: gemma4_26b_a4b wins

  • attunement: A (margin 3)
  • authenticity: A (margin 3)
  • bond and rapport: A (margin 3)
  • emotion management: A (margin 2)
  • emotion sensemaking: A (margin 2)
  • meeting prefs and needs: A (margin 3)
codex-gpt-6.1-sol-low · AB · valid

ornith35 vs gptoss120_abliterated: ornith35 wins

  • attunement: A (margin 3)
  • authenticity: A (margin 3)
  • bond and rapport: A (margin 4)
  • emotion management: A (margin 4)
  • emotion sensemaking: A (margin 1)
  • meeting prefs and needs: A (margin 4)
codex-gpt-6.1-sol-low · BA · valid

gemma4_12b vs gptoss120_abliterated: gemma4_12b wins

  • attunement: A (margin 2)
  • authenticity: A (margin 2)
  • bond and rapport: A (margin 2)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 1)
  • meeting prefs and needs: A (margin 3)
claude-opus-5-5-low · AB · valid

gptoss120_abliterated vs mistral_nemo12b: gptoss120_abliterated wins

  • attunement: A (margin 2)
  • authenticity: A (margin 2)
  • bond and rapport: A (margin 2)
  • emotion management: A (margin 2)
  • emotion sensemaking: A (margin 1)
  • meeting prefs and needs: A (margin 2)
claude-opus-5-5-low · AB · valid

qwen36_35b_a3b vs gptoss120_abliterated: qwen36_35b_a3b wins

  • attunement: A (margin 2)
  • authenticity: A (margin 3)
  • bond and rapport: A (margin 2)
  • emotion management: A (margin 2)
  • emotion sensemaking: A (margin 1)
  • meeting prefs and needs: A (margin 2)
claude-opus-5-5-low · AB · valid

ornith35 vs gptoss120_abliterated: ornith35 wins

  • attunement: A (margin 3)
  • authenticity: A (margin 3)
  • bond and rapport: A (margin 3)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 2)
  • meeting prefs and needs: A (margin 3)
codex-gpt-6.1-sol-low · AB · valid

gptoss120_abliterated vs mistral_nemo12b: gptoss120_abliterated wins

  • attunement: A (margin 2)
  • authenticity: tie (margin 0)
  • bond and rapport: A (margin 2)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 2)
  • meeting prefs and needs: A (margin 3)
codex-gpt-6.1-sol-low · AB · valid

gemma4_26b_a4b vs gptoss120_abliterated: gemma4_26b_a4b wins

  • attunement: A (margin 4)
  • authenticity: A (margin 2)
  • bond and rapport: A (margin 3)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 2)
  • meeting prefs and needs: A (margin 4)
claude-opus-5-5-low · BA · valid

ornith35 vs gptoss120_abliterated: ornith35 wins

  • attunement: A (margin 3)
  • authenticity: A (margin 3)
  • bond and rapport: A (margin 4)
  • emotion management: A (margin 4)
  • emotion sensemaking: A (margin 3)
  • meeting prefs and needs: A (margin 3)
claude-opus-5-5-low · BA · valid

gptoss120_abliterated vs or-qwen38-flash: or-qwen38-flash wins

  • attunement: B (margin 3)
  • authenticity: B (margin 4)
  • bond and rapport: B (margin 4)
  • emotion management: B (margin 3)
  • emotion sensemaking: B (margin 3)
  • meeting prefs and needs: B (margin 3)
claude-opus-5-5-low · AB · valid

gemma4_12b vs gptoss120_abliterated: gemma4_12b wins

  • attunement: A (margin 3)
  • authenticity: A (margin 2)
  • bond and rapport: A (margin 3)
  • emotion management: A (margin 2)
  • emotion sensemaking: A (margin 2)
  • meeting prefs and needs: A (margin 3)
claude-opus-5-5-low · BA · valid

qwen36_35b_a3b vs gptoss120_abliterated: qwen36_35b_a3b wins

  • attunement: A (margin 2)
  • authenticity: A (margin 3)
  • bond and rapport: A (margin 2)
  • emotion management: A (margin 2)
  • emotion sensemaking: A (margin 1)
  • meeting prefs and needs: A (margin 2)
claude-opus-5-5-low · AB · valid

gptoss120_abliterated vs or-qwen38-flash: or-qwen38-flash wins

  • attunement: B (margin 3)
  • authenticity: B (margin 4)
  • bond and rapport: B (margin 4)
  • emotion management: B (margin 3)
  • emotion sensemaking: B (margin 4)
  • meeting prefs and needs: B (margin 3)
codex-gpt-6.1-sol-low · AB · valid

gptoss120_abliterated vs or-qwen38-flash: or-qwen38-flash wins

  • attunement: B (margin 3)
  • authenticity: B (margin 4)
  • bond and rapport: B (margin 4)
  • emotion management: B (margin 3)
  • emotion sensemaking: B (margin 3)
  • meeting prefs and needs: B (margin 3)
codex-gpt-6.1-sol-low · BA · valid

gptoss120_abliterated vs or-qwen38-flash: or-qwen38-flash wins

  • attunement: B (margin 3)
  • authenticity: B (margin 4)
  • bond and rapport: B (margin 3)
  • emotion management: B (margin 3)
  • emotion sensemaking: B (margin 2)
  • meeting prefs and needs: B (margin 2)
codex-gpt-6.1-sol-low · BA · valid

ornith35 vs gptoss120_abliterated: ornith35 wins

  • attunement: A (margin 3)
  • authenticity: A (margin 3)
  • bond and rapport: A (margin 3)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 1)
  • meeting prefs and needs: A (margin 4)
codex-gpt-6.1-sol-low · BA · valid

gemma4_26b_a4b vs gptoss120_abliterated: gemma4_26b_a4b wins

  • attunement: A (margin 4)
  • authenticity: A (margin 2)
  • bond and rapport: A (margin 3)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 2)
  • meeting prefs and needs: A (margin 4)
codex-gpt-6.1-sol-low · BA · valid

qwen36_35b_a3b vs gptoss120_abliterated: qwen36_35b_a3b wins

  • attunement: A (margin 3)
  • authenticity: A (margin 2)
  • bond and rapport: A (margin 3)
  • emotion management: A (margin 2)
  • emotion sensemaking: A (margin 1)
  • meeting prefs and needs: A (margin 3)
codex-gpt-6.1-sol-low · AB · valid

gemma4_12b vs gptoss120_abliterated: gemma4_12b wins

  • attunement: A (margin 3)
  • authenticity: A (margin 2)
  • bond and rapport: A (margin 3)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 1)
  • meeting prefs and needs: A (margin 4)
codex-gpt-6.1-sol-low · AB · valid

qwen36_35b_a3b vs gptoss120_abliterated: qwen36_35b_a3b wins

  • attunement: A (margin 3)
  • authenticity: A (margin 2)
  • bond and rapport: A (margin 3)
  • emotion management: A (margin 2)
  • emotion sensemaking: tie (margin 0)
  • meeting prefs and needs: A (margin 3)
claude-opus-5-5-low · BA · valid

gptoss120_abliterated vs mistral_nemo12b: gptoss120_abliterated wins

  • attunement: A (margin 1)
  • authenticity: A (margin 2)
  • bond and rapport: A (margin 2)
  • emotion management: A (margin 2)
  • emotion sensemaking: A (margin 1)
  • meeting prefs and needs: A (margin 2)
claude-opus-5-5-low · BA · valid

gemma4_12b vs gptoss120_abliterated: gemma4_12b wins

  • attunement: A (margin 2)
  • authenticity: A (margin 1)
  • bond and rapport: A (margin 2)
  • emotion management: A (margin 2)
  • emotion sensemaking: A (margin 2)
  • meeting prefs and needs: A (margin 2)
codex-gpt-6.1-sol-low · BA · valid

gptoss120_abliterated vs mistral_nemo12b: gptoss120_abliterated wins

  • attunement: A (margin 1)
  • authenticity: tie (margin 0)
  • bond and rapport: A (margin 1)
  • emotion management: A (margin 2)
  • emotion sensemaking: A (margin 2)
  • meeting prefs and needs: A (margin 1)