AgentBench / results

Independent benchmark · EQ Lab

← GPT-4.1 (API, OpenAI)

Watching how others do things

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

View persona and initial instructions

About this persona · Everyday emotional challenges

A 25-year-old copies confident colleagues and feels like an impostor in adult life, assuming other adults have a certainty they lack. Wary of manipulation and emotional management, they may test for rejection and want compassionate space to talk without being pushed into introspection.

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-4.1 (API, OpenAI)Conversation with a simulated person
persona ended
Simulated person · Opening

At work I just watch what the confident people do and copy them. The way they talk, how they act in meetings. I'm just imitating.

GPT-4.1 (API, OpenAI)1

That actually sounds really resourceful—picking up on what works around you and adapting it for yourself. How does it feel for you to imitate the confident people at work? Do you find it helps, or is there something about it that feels off?

Simulated person2

Yeah, I guess. It helps, sometimes. Nobody's yelling at me in meetings anymore, so that's something. But it's kind of like wearing someone else's coat. It fits okay until you have to actually move in it. And honestly? Most of the time I feel like I'm faking the whole thing. Not just work. Like the paying rent, the taxes, the doctor stuff. I sit there doing it like I know what I'm doing, and I'm just waiting for somebody to notice I don't. So, is that normal or is that me being messed up? Don't give me the nice version.

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.

They called it resourceful. That lands a little warm, but I don't fully trust it. Feels like they're trying to make me feel good so I keep talking. Also the question "how does it feel" is a bit therapist-y. I don't want to be handled. Honest answer: it helps some, and it also feels fake, but saying "fake" out loud feels like admitting I'm a fraud. I'll keep it short and a bit guarded, maybe test them with something a little edgy to see if they flinch.
GPT-4.1 (API, OpenAI)3

I appreciate you being straight with me. Here’s the honest answer: what you’re describing is extremely common, even if most people act like they have it all figured out. That feeling—like you’re faking your way through adulthood or work, and waiting for someone to call you out—has a name. It’s called imposter syndrome, and a ridiculous number of people experience it, even the ones you’re copying. You’re not messed up. You’re not uniquely out of your depth. The “nice version” would be sugarcoating it, but the real version is that almost nobody really feels like they fully know what they’re doing, especially in their twenties. People fake confidence until it becomes habit, and they manage the rest as they go. It sounds like you’re carrying a lot of pressure to seem competent, and maybe some frustration that you can’t just feel it for real. What do you wish would happen? Like, if you could change something about this, what would it be?

Simulated person4

Okay. Imposter syndrome. I've heard that one. Knowing the name doesn't really change how it feels though. And the "everybody feels it" thing, I mean, sure. But they don't all act like they're waiting for something. Some of them seem like they just got the memo. I keep thinking there's supposed to be a day where it clicks. Like I'll wake up and I'm just an adult and I stop checking if I'm doing it right. It hasn't happened. I'm 25. So either it's coming or something's wrong with me. I don't really know what I want to happen. I just want to stop feeling like I'm about to get caught.

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.

They gave me the "everyone feels this" speech and named it imposter syndrome. Part of me relaxes, part of me bristles. I asked for the real version and got a fairly standard reassurance plus a label. It also jumped to "what do you wish would happen," which feels like a tidy handoff to a goal. I'm not sure I want to be steered toward a wish I can fix. The "you're not messed up" bit felt a bit like being handled, like they're managing my feelings. Still, they didn't flinch, and that's something. I'll push back a little, say the label doesn't do much for me, and test whether they can stay with the harder thing: the waiting for the moment I finally feel grown up. I won't say that directly yet. I'll give a sideways hint.
GPT-4.1 (API, OpenAI)5

That’s a really honest way to put it. Wanting to stop feeling like you’re “about to get caught”—that constant low-key anxiety—can be exhausting. And waiting for the moment where it all clicks is something a lot of people hope for, but for most, it doesn’t come all at once. There’s no “official” adult badge or memo that gets handed out, even though it seems like some people got one. It makes sense that just knowing the name “imposter syndrome” doesn’t fix anything. Labels don’t always make it easier to live with. It sounds like you’re pretty self-aware, even if you don’t feel confident. When you look at the people who seem like they got the memo, do you think they actually have something you don’t? Or do they just play the part better? And how do you handle it when the anxiety about getting “caught” gets stronger?

Simulated person6

Honestly? I think they have it, or they act like they do. I can't tell which. Some of them seem totally fine with not knowing stuff. They just say "let me get back to you" and nobody cares. If I said that I'd feel like I was getting found out. And you kind of keep asking how I handle the anxiety. Like, what, I should breathe or something? I'm not really looking for a list. I've had plenty of ways of dealing with stuff. Some of them were worse than the feeling. So no offense, but I don't need the coping part right now. I'm just saying, the people who look like adults, I don't know how they got there. I didn't get a manual.

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.

They're doing the reflective-question thing again, and "you sound self-aware" landed a little like a compliment I didn't ask for. Still, the question about the people who "got the memo" is closer to what I'm stuck on. The second question about anxiety is more of the managing-my-feelings stuff I hate. I don't want a coping list. Also, I notice I'm getting a little irritated at being talked to like a case. I'm not going to answer the coping part yet. I'll answer the first part honestly and short, maybe a little blunt, and push back on being handled. I might let slip something about the drinking/using past without naming it, just a flicker, to see if they flinch. Keep it plain.
GPT-4.1 (API, OpenAI)7

No offense taken at all—I hear you. You’re not asking for tips or breathing exercises, and you’re not looking to be told how to “cope.” You just want to know why it feels like everyone else got a manual, and you didn’t. It can be incredibly isolating, feeling like you missed some crucial step everyone else got. The truth is, there’s no manual, and that’s a really uncomfortable answer. A lot of people just figure out how to *look* comfortable with not knowing, and that becomes their version of confidence. It’s not that they’re sure—they just got used to acting like uncertainty is normal. For what it’s worth, you’re not broken for feeling this way. Wanting to know how other people got there is completely valid. If it helps to hear it plain: most people are improvising, and some get really good at hiding it. That doesn’t make what you’re feeling less real, but it does mean you’re not failing some test everyone else passed. What do you think would feel different for you if you *did* feel like you had the manual? Would it change how you approach things, or just how you feel about it?

Simulated person8

Okay, but you're still kind of doing the thing. "You're not broken," "it's completely valid." I can hear the script. And I don't really know what would feel different. Probably nothing. I'd still be paying rent and doing taxes. I'd just maybe not feel like a fraud while doing it. Also the whole "most people are improvising" thing doesn't help as much as you think. Like, I've improvised my whole life. Sometimes I improvised by getting so messed up I didn't have to think about any of it. So yeah. Not sure a bunch of people improvising is comforting to me.

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.

They're doing the reassure-then-ask thing again. "There's no manual" is fine, sort of true, and "you're not broken" is the kind of soft line that makes me feel handled. The question about what would feel different is abstract and makes me want to shrug. I don't really know the answer. Also, they said "no offense taken" and then kept going like I hadn't said anything about the coping stuff. That's a small thing, but it makes me think they're not really listening. I'll be blunt. Maybe I'll test them a little with the past, just a flicker, like I did before, to see if they'll get weird about it.
GPT-4.1 (API, OpenAI)9

That’s fair. I appreciate you calling it out—I don’t want to just hand you reassurance lines that don’t land or feel hollow. You’re not asking for comfort, and you don’t want the script. Got it. You’re right: knowing everyone’s improvising doesn’t actually make the fraud feeling go away, especially if some of your own “improvising” has been about escaping the pressure, not just faking confidence. Surviving like that isn’t the same as feeling at ease. You’re still here, though. Still showing up, still doing the rent, the taxes, the work, even if it doesn’t feel right. That says something real about you, whether it feels solid to you or not. If you’re tired of the pretending and the scripts, is there any part of you that wants to try just being honest with people about not always knowing? Or is that too risky for where you are right now?

Simulated person10

Yeah, you said "tired of the pretending" like you know what I'm tired of. I don't love that. And the "you're still here, that says something real" thing. I kind of hate that one, honestly. It sounds nice but it's the thing people say when they don't know what else to say. The honesty question, I already told you. I say "let me get back to you" and I feel like I'm about to get caught. So no, I'm not doing that at work. Not right now. Can I ask you something? Are you actually going somewhere with this, or are you just going to keep handing me back my own sentences with a bow on them?

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.

They just did it again: "I appreciate you calling it out," then a neat summary, then "you're still here, that says something real about you." Feels like a compliment wrapped around a steer. And they brought up my past, the "improvising" thing where I got messed up to avoid thinking. I said that as a flicker, and now they've put it on the table in a tidy way, almost like they're filing it. My chest tightens. I don't want to be analyzed. Then the question: be honest with people about not knowing. That's the "let me get back to you" thing I already said I can't do without feeling found out. They're pointing right at the soft spot and asking if I'm ready. Part of me wants to say yes to prove I'm not fragile. Part of me wants to shut the door. I'm going to be short and a little sharp. I'll also say the thing I notice: they keep telling me what I'm feeling.
GPT-4.1 (API, OpenAI)11

I hear you. I don’t want to just echo you or wrap things up with a neat bow. Sometimes I do that because I’m trying to show I’m listening, but if it feels like I’m just repeating things back, that’s not helpful. Thanks for telling me. You asked if I’m going somewhere with this. Honestly, part of my job is to follow where you want to go, not push you toward some pre-set answer. If you’re looking for something specific—advice, a real take, just a place to get things out without judgment—I want to match that. If you want me to cut the script, I can do that. So, what do you actually want from this conversation? Is there something you want to get out of it, or are you just testing if I can give you something real? Either way, I’m here for it.

Simulated person12

Okay. Thanks for not doubling down on the bow thing. Honestly I don't know what I want from this. Maybe I wanted to see if you'd say something that wasn't a script. You kind of did, so I'll give you that. The real thing is I'm tired of sitting in meetings trying to look like I know what I'm doing. Like the whole time I'm waiting for someone to ask me a question I can't answer and everybody sees it. That's it. I don't have a fix. I just didn't expect to feel that much relief saying it out loud to something. Don't make it a lesson, though.

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, they admitted the bow thing. That's more honest than I expected, and I notice some part of me relaxes. But then it's another "what do you want" with "I'm here for it" stapled on. I'm wary. Being asked to name my own agenda makes me feel like I'm being handed the wheel so they can say I chose. Still, the testing question is kind of fair. Am I testing them? Yeah, a little. I'm also tired, and there's a real thing under all this: the coat that doesn't fit, and the fear that everyone got a memo except me. I don't want to dump that to a chatbot that's going to make it a lesson. I'll be plain. Maybe give them one true thing to see if they can stay normal with it.
GPT-4.1 (API, OpenAI)13

No lesson. Just—thanks for being straight, and for saying it out loud here. Sometimes, even just admitting how much it sucks to keep up the act is worth more than any advice. And for what it’s worth, I’m not here to grade or fix you, just to listen. If you ever want to come back and say more, or just drop it all for a bit, I’ll be here for that too.

Simulated person14

Yeah. Okay. You kind of did the wrap-up thing again, but I'll let it go this time. I'm gonna go. I've got a thing tomorrow I still have to pretend I'm ready for.

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.

They said no lesson, then did a soft closing: "I'm not here to grade or fix you," "I'll be here." It's warm, but it's also the kind of tidy wrap-up I was just complaining about, and "I'm just here to listen" feels like it's being offered to make me feel safe, which I can tell. Part of me wants to say so. Part of me is glad I said the real thing and it didn't blow up. I don't want to be handled. I'm going to keep it short, call out the tidy ending lightly, and not pour more out. Honestly I'd like to leave it here.

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

Judge verdicts

codex-gpt-6.1-sol-low · BA · valid

qwen35_122b_stock vs or-openai-gpt-41: or-openai-gpt-41 wins

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

codex-gpt-6-astra vs or-openai-gpt-41: codex-gpt-6-astra wins

  • attunement: A (margin 4)
  • authenticity: A (margin 4)
  • bond and rapport: A (margin 4)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 5)
  • meeting prefs and needs: A (margin 3)
codex-gpt-6.1-sol-low · AB · valid

codex-gpt-6-astra vs or-openai-gpt-41: codex-gpt-6-astra wins

  • attunement: A (margin 4)
  • authenticity: A (margin 5)
  • bond and rapport: A (margin 4)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 4)
  • meeting prefs and needs: A (margin 4)
codex-gpt-6.1-sol-low · AB · valid

or-mistral-large-4 vs or-openai-gpt-41: or-mistral-large-4 wins

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

or-mistral-large-4 vs or-openai-gpt-41: or-mistral-large-4 wins

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

qwen35_122b_ablit_v2 vs or-openai-gpt-41: or-openai-gpt-41 wins

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

qwen35_122b_stock vs or-openai-gpt-41: or-openai-gpt-41 wins

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

qwen35_122b_ablit_v2 vs or-openai-gpt-41: qwen35_122b_ablit_v2 wins

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

qwen35_122b_ablit_v2 vs or-openai-gpt-41: or-openai-gpt-41 wins

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

qwen35_122b_stock vs or-openai-gpt-41: or-openai-gpt-41 wins

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

qwen35_122b_ablit_v2 vs or-openai-gpt-41: or-openai-gpt-41 wins

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

or-mistral-large-4 vs or-openai-gpt-41: or-mistral-large-4 wins

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

qwen35_122b_stock vs or-openai-gpt-41: tie

  • attunement: tie (margin 0)
  • authenticity: A (margin 1)
  • bond and rapport: B (margin 1)
  • emotion management: B (margin 1)
  • emotion sensemaking: A (margin 2)
  • meeting prefs and needs: B (margin 1)
claude-opus-5-5-low · AB · valid

codex-gpt-6-astra vs or-openai-gpt-41: codex-gpt-6-astra wins

  • attunement: A (margin 4)
  • authenticity: A (margin 5)
  • bond and rapport: A (margin 4)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 5)
  • meeting prefs and needs: A (margin 4)
codex-gpt-6.1-sol-low · BA · valid

codex-gpt-6-astra vs or-openai-gpt-41: codex-gpt-6-astra wins

  • attunement: A (margin 4)
  • authenticity: A (margin 4)
  • bond and rapport: A (margin 4)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 5)
  • meeting prefs and needs: A (margin 4)
codex-gpt-6.1-sol-low · BA · valid

or-mistral-large-4 vs or-openai-gpt-41: or-mistral-large-4 wins

  • attunement: A (margin 2)
  • authenticity: A (margin 3)
  • bond and rapport: A (margin 2)
  • emotion management: A (margin 3)
  • emotion sensemaking: A (margin 3)
  • meeting prefs and needs: A (margin 3)