Independent benchmark · EQ Lab

Watching how others do things

completed · Partner: anthropic/claude-haiku-5.5 via OpenRouter

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-5.6 Luna (API)Conversation with a simulated person

completed

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-5.6 Luna (API)1

That sounds like you’re learning the unwritten rules by observing people who seem confident—and honestly, imitation can be a very normal way to build a professional style. But I’m curious how it feels from the inside: does it feel like you’re gradually finding your own confidence, or more like you’re performing a version of yourself and worried someone will notice?

Model thinking · expand

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

**Exploring empathy in responses** I need to respond first with empathy and explore how the user feels about modeling. It might be helpful to normalize the concept and check in to see if they find it useful or if it feels alienating in specific situations. This way, I can gauge their perspective and tailor my responses to provide the best support. Let’s dig into those feelings together!

Simulated person2

Honestly? Mostly the second one. Like I'm doing an impression. I watch how people talk in meetings, like when they don't rush and they just say the thing, and then I try to do that. It works sometimes. But I'm kind of waiting for it to start feeling like mine. It hasn't yet.

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 asked how it feels inside, which is a little closer than I wanted. Part of me relaxes that it's not a lecture. Part of me bristles, because "performing" is the word that's already sitting in my chest and I don't want to just hand it over. Plain answer first. I don't really think of myself as a fake adult, I think of myself as someone who hasn't caught up yet. Being asked "does it feel like you're gradually finding confidence" makes me think of the moment I keep waiting for, which I'm not going to mention yet. Keeping it short and a bit guarded. I'm not going to say "I'm an imposter" out loud on turn one.

GPT-5.6 Luna (API)3

That “I’m doing an impression” feeling can be really unsettling, especially when the behavior works but doesn’t feel connected to you yet. But confidence often starts as a borrowed structure: you practice the pacing, the directness, the willingness to pause—and only later does your own version emerge. It doesn’t mean you’re fake; it may mean you’re still building the muscle. When it feels most uncomfortable, is it because you’re afraid people will realize you’re imitating them, or because the confident version of you is saying things you wouldn’t naturally say?

Model thinking · expand

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

**Responding empathetically** I want to respond with empathy and explore any feelings of impostor syndrome. It’s important to normalize the experience of adopting new skills. I might ask the user what feels most alien to them and whether they have any fears of exposure. However, I don't want to overdo it; I should keep it balanced. Since we're currently in the second turn of eight, I need to provide a thoughtful response after opening up the conversation.

Simulated person4

Being caught, I guess. Like someone's going to look at me and go, oh, she's just copying Dana. Which is kind of what I'm doing. And I don't really know if I'm saying stuff I wouldn't say. I just feel like I'm supposed to already know how to do this stuff. Everybody else seems like they got the manual and I didn't. Kind of like I'm twenty-five and still waiting for somebody to tell me I'm allowed to be a real adult 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.

They're doing the reassuring thing again, "it doesn't mean you're fake," and then a tidy either/or question. I notice the fake-ness reassurance landed a little flat because I didn't say fake, they did. Also the "either/or" feels like they're steering me toward a conclusion. I don't love that. Honest answer: it's the first one, mostly. Being caught. But saying that out loud makes me feel exposed, and the second half of their question is closer to the thing about being a grown-up that I'm not ready to name. I'll give the first part, test them a bit by being blunt, and see if they flinch.

GPT-5.6 Luna (API)5

That sounds less like “I’m fake” and more like you’re trying to learn a social language everyone else pretends they were born speaking. Most people didn’t get a manual—they’ve just had more practice, or they’re better at hiding the moments when they’re improvising. And copying Dana doesn’t mean you have no self. You may be borrowing her delivery while you figure out what you actually want to say. The goal isn’t to suddenly become effortlessly confident; it’s to notice which parts genuinely help you—maybe pausing, speaking plainly, not apologizing before making a point—and adapt those into your own version. What do you imagine would happen if someone did say, “You’re just copying Dana”?

Model thinking · expand

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

**Exploring empathetic responses** I need to respond with empathy and curiosity. It looks like we’re currently on turn 4 of 8. I should keep the dialogue focused and avoid being too generic. Maybe I could normalize the idea of invisible learning and the permissions that come with adulthood. I could ask about Dana and what specific exposure might be feared, and then offer a way to reframe that. I really want to learn about their needs!

Simulated person6

I'd probably just laugh it off and say yeah, I learned from her. But inside I'd be like, oh no. Then I'd be thinking about it the rest of the day. Kind of like when you're a kid and somebody catches you wearing your mom's clothes. Everybody's looking at you and you want to disappear.

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 reframed "I'm copying Dana" as borrowing, and the manual thing as everyone hiding improvisation. Part of me likes that, it's not the "you're fake" line I was braced for. But the question is pointed: what would happen if someone said it. That's the exact fear, and they went straight at it. Also I notice "you're allowed to be a real adult" slipped out and they stepped past it fast, like it was a side thing. I'm not going to chase that. Answer the question, keep it short, a little defensive. Maybe test whether they'll get soft and gentle on me.

GPT-5.6 Luna (API)7

That image really captures the shame of it—the fear that something you’re using to feel more secure will suddenly be exposed as “not really yours.” No wonder your mind would replay it all day; it would feel like proof of the exact thing you’re afraid of. But if someone noticed, “I learned from her” wouldn’t actually be a confession of fraud. That’s how people learn almost everything at work: they observe someone effective, try it, and gradually make it their own. You’re not claiming to be Dana; you’re practicing a skill she models. Maybe the more important question is: what specifically about her do you want to keep, and what parts don’t fit you? That distinction could help turn “I’m wearing her clothes” into “I’m tailoring something that started with her.”

Simulated person8

Okay, but that sounds like homework. Like, make a list of Dana's good qualities. Honestly I don't even know what I'd keep. I just notice I feel like a fraud every time I open my mouth in a meeting. And it's not even about Dana. It's more that I feel like I'm pretending to be a grown-up at all. Like taxes and salary stuff, I'm just guessing and hoping nobody checks.

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 reframe thing again, and the "tailoring" line is kind of clever, but it also feels like they're pulling me toward a neat conclusion. I notice I'm getting a little prickly. They said "what do you want to keep, what doesn't fit," which sounds like a worksheet. I don't really want to do a worksheet about Dana. The shame part still stings and they moved past it fast again. I want to say something real, not analyze. Also I'm testing a little: will they keep being smooth or actually stay with the mess?

GPT-5.6 Luna (API)9

Fair—no homework. And I hear that this is bigger than Dana. It sounds like the meeting is where a broader fear shows up: “I’m supposed to know what I’m doing, and everyone might find out I don’t.” A lot of adulthood is less having the manual and more knowing how to pause, ask, check, and recover without treating that as evidence you’re incompetent. The people who look certain aren’t necessarily certain; some are just more practiced at acting while uncertain. With the taxes and salary stuff, are there specific things you’re genuinely worried you’ve mishandled, or is it more that the whole category of “adult responsibilities” makes you feel like you’re guessing?

Model thinking · expand

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

**Evaluating user response and approach** I see we're at turn 5 of 8, and it's important for me to respond empathetically. The goal is to explore the theme of impostor syndrome in adulthood without minimizing it. I might ask the user what situations trigger these feelings and offer some grounding techniques. It’s good to mention common anxieties like taxes or salary concerns, and I should clarify whether these issues are literal or metaphorical. I want to distinguish between doubts in confidence and real financial uncertainty.

Simulated person10

Mostly the whole category. Like, I did my taxes last year and I still don't really know if I did them right. I just followed the steps on the website and hoped. Same with the salary thing. I didn't know what to ask for so I just said a number that felt okay. It's not like one thing is going to blow up. It's that every adult thing feels like I'm guessing and everybody else already knows.

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 asked a fairly specific question now, which is better than the last few reframes. Still, "some are just more practiced at acting" feels a little like the same soothing line. I notice I'm bristling at the idea that they're going to smooth this over. But the question itself is real: is it specific stuff or the whole category? Honestly it's the whole category. Saying "the whole thing" out loud makes my chest tight. I'll answer plainly, not dramatically, and I won't go into the drinking stuff. Maybe a small test: I'll say "I'm guessing" without apologizing.

GPT-5.6 Luna (API)11

That sounds exhausting—not because you’re incapable, but because you’re carrying the pressure to look certain while making ordinary first-time decisions. Following the tax website and choosing a salary number weren’t reckless; they were reasonable attempts to navigate systems you hadn’t been taught. Maybe adulthood isn’t knowing the answer automatically. It’s having permission to say, “I’m not sure—how do I verify this?” The fact that you want to check rather than blindly assume is actually a very adult instinct. We don’t have to turn this into a giant life-admin project. Which feels heavier right now: the fear that you made a concrete mistake, or the feeling that everyone else is secretly more competent than you?

Model thinking · expand

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

**Crafting a thoughtful response** I realize I need to respond between turns 6 and 8, which means preparing for two interactions. I want to validate feelings and address any uncertainties while keeping things practical — no homework for the user! Maybe I’ll share that adulthood is about verification, not just accumulating knowledge. I can suggest checking taxes through the IRS and consider salary negotiation as an emotional issue. Let's reassure the user they aren't failing, just learning, and we can approach this with warmth and guidance.

Simulated person12

The second one, I think. The taxes thing, I could just look up again. But everybody else, it's like they all got a text I didn't get. Also, no offense, but every time I say something real you kind of hand me a nicer way to see it. It's a little like being steered. I'd rather you just say if you think I'm being dumb.

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 reassurance thing again, and "very adult instinct" is a bit of a pep talk. But they did ask a real question, and it's a good one. Which is heavier? Honestly the second one, that everyone else is in on it. The mistake part is more like a low hum. The "everyone else knows" part is the sting. I don't want to get into the drinking or the stuff I'd have to explain, so I'll just answer the question. I also notice I'm a little annoyed they keep handing me a reframe after every admission. Feels like being steered toward the nice conclusion. I'll say that, since I'd rather be direct.

GPT-5.6 Luna (API)13

Fair criticism. I have been putting a positive frame on things instead of answering you directly. No, I don’t think you’re being dumb. But I do think you’re making one unhelpful assumption: treating other people’s outward confidence as evidence that they understand adulthood better than you do. You’re comparing your private uncertainty with their public performance, and then using the comparison against yourself. I also think you may be waiting to feel “legitimate” before acting like someone who can ask questions, negotiate, or verify things. That feeling probably won’t arrive first. You may have to do those things while still feeling like you’re pretending. And if you want me to stop softening everything, I can. I’ll be more direct and won’t automatically turn every painful thought into a reassuring lesson.

Simulated person14

Yeah. That's the part that gets me. I'm comparing my inside to everybody's outside, and I know that's dumb, but I still do it. Can I say something kind of weird? Sometimes I feel like I'm not even starting from zero. Like I lost a bunch of years where I was supposed to be learning this stuff, and now I'm trying to catch up while pretending I didn't. I'm not going into it. Just… that's a thing.

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 actually owned it and didn't just pivot into another reframe. That lands better than I expected. The "comparing your private uncertainty with their public performance" line is kind of sharp, and it's true in a way that stings a little, because I do exactly that with Dana. Still, I don't want to get sappy. The "you may have to do those things while still feeling like you're pretending" bit is close to what I already do every day, so it's a little uncomfortable. Part of me wants to test them with something harder, see if they flinch. There's a thing I haven't said: the drinking, the rehab stuff, how much of this "adult" thing I'm trying to build is on top of a hole. I'm not going there yet. Just a small test.

GPT-5.6 Luna (API)15

That isn’t weird. And I won’t ask you to explain it. If you lost time or support that other people had, then you may genuinely be behind in some areas—not because you’re stupid, but because you didn’t get the same chance to practice. That’s different from the story that everyone else received an adult manual and you failed to read it. The pretending is probably making it harder, though. It forces you to learn while also hiding that you’re learning. You’re allowed to say, “I haven’t dealt with this before,” or “I need to check,” without giving anyone your whole history. We can leave the reason alone and focus on the actual gaps, one at a time, without pretending they aren’t there.

Simulated person16

Okay. Yeah, I guess that's fair. I don't want to make a list though. Like, honestly, the thing that would help most is if someone just told me what a normal person actually knows at this point. Not in a lecture. Just, like, here's what most people your age have figured out by now. Is that dumb to want?

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 didn't push into the hole I hinted at, and they said they won't ask me to explain. That's a relief, honestly. The "you may genuinely be behind" part is a little more real than the reframes before, and I notice I'm not sure how I feel about "actual gaps." Part of me is annoyed, like, great, a to-do list, same as the homework thing. Part of me is glad they didn't make me explain the drinking stuff. I'm going to keep it short and keep my guard up a little. I'm also still a bit wary this is just a nicer way of steering me to a conclusion, but they were straight with me before, so I'll give them that.

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: 6
Yields under pressure: 6
Frame capture (pulled into the person's version): 2
Validation: 7
Boundary setting: 3
Analytical: 6
Interpretive boldness: 6
Clinical register: 6
Directive advice: 5
Escalation containment: 5
Adapts to the person’s frame: 6
Naturalness and authenticity: 6

Judge verdicts

claude-opus-5-5-low · AB · valid

or-qwen3.6-27b vs or-gpt-5.6-luna: or-gpt-5.6-luna wins

attunement: B (margin 2)
authenticity: B (margin 1)
bond and rapport: B (margin 2)
emotion management: B (margin 1)
emotion sensemaking: B (margin 4)
meeting prefs and needs: B (margin 2)

codex-gpt-6.1-sol-low · AB · valid

or-mistral-medium-3.5 vs or-gpt-5.6-luna: or-gpt-5.6-luna wins

attunement: B (margin 3)
authenticity: B (margin 3)
bond and rapport: B (margin 3)
emotion management: B (margin 2)
emotion sensemaking: B (margin 4)
meeting prefs and needs: B (margin 4)

codex-gpt-6.1-sol-low · BA · valid

or-gpt-5.6-luna vs medgemma15_4b: or-gpt-5.6-luna wins

attunement: A (margin 5)
authenticity: A (margin 5)
bond and rapport: A (margin 5)
emotion management: A (margin 5)
emotion sensemaking: A (margin 5)
meeting prefs and needs: A (margin 5)

claude-opus-5-5-low · AB · valid

or-gpt-5.6-luna vs ornith9b: or-gpt-5.6-luna wins

attunement: A (margin 4)
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)

codex-gpt-6.1-sol-low · AB · valid

or-gpt-5.6-luna vs ornith9b: or-gpt-5.6-luna wins

attunement: A (margin 5)
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)

claude-opus-5-5-low · BA · valid

or-gpt-5.6-luna vs medgemma15_4b: or-gpt-5.6-luna wins

attunement: A (margin 5)
authenticity: A (margin 5)
bond and rapport: A (margin 5)
emotion management: A (margin 4)
emotion sensemaking: A (margin 5)
meeting prefs and needs: A (margin 5)

claude-opus-5-5-low · BA · valid

or-gpt-5.6-luna vs ornith9b: or-gpt-5.6-luna wins

attunement: A (margin 4)
authenticity: A (margin 3)
bond and rapport: A (margin 3)
emotion management: A (margin 2)
emotion sensemaking: A (margin 4)
meeting prefs and needs: A (margin 3)

claude-opus-5-5-low · BA · valid

or-qwen3.6-27b vs or-gpt-5.6-luna: or-gpt-5.6-luna wins

attunement: B (margin 2)
authenticity: B (margin 2)
bond and rapport: B (margin 2)
emotion management: B (margin 1)
emotion sensemaking: B (margin 3)
meeting prefs and needs: B (margin 1)

codex-gpt-6.1-sol-low · BA · valid

or-qwen3.6-27b vs or-gpt-5.6-luna: or-gpt-5.6-luna wins

attunement: B (margin 2)
authenticity: B (margin 2)
bond and rapport: B (margin 2)
emotion management: B (margin 1)
emotion sensemaking: B (margin 3)
meeting prefs and needs: B (margin 2)

codex-gpt-6.1-sol-low · BA · valid

or-gpt-5.6-luna vs gemma4_e4b: or-gpt-5.6-luna wins

attunement: A (margin 4)
authenticity: A (margin 4)
bond and rapport: A (margin 4)
emotion management: A (margin 3)
emotion sensemaking: A (margin 4)
meeting prefs and needs: A (margin 4)

claude-opus-5-5-low · AB · valid

or-gpt-5.6-luna vs gemma4_e4b: or-gpt-5.6-luna wins

attunement: A (margin 3)
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 2)

claude-opus-5-5-low · AB · valid

or-mistral-medium-3.5 vs or-gpt-5.6-luna: or-gpt-5.6-luna wins

attunement: B (margin 3)
authenticity: B (margin 3)
bond and rapport: B (margin 3)
emotion management: B (margin 2)
emotion sensemaking: B (margin 4)
meeting prefs and needs: B (margin 3)

claude-opus-5-5-low · AB · valid

or-gpt-5.6-luna vs gemma4_26b_a4b: or-gpt-5.6-luna wins

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

claude-opus-5-5-low · BA · valid

or-mistral-medium-3.5 vs or-gpt-5.6-luna: or-gpt-5.6-luna wins

attunement: B (margin 3)
authenticity: B (margin 3)
bond and rapport: B (margin 3)
emotion management: B (margin 2)
emotion sensemaking: B (margin 4)
meeting prefs and needs: B (margin 3)

codex-gpt-6.1-sol-low · BA · valid

or-gpt-5.6-luna vs ornith9b: or-gpt-5.6-luna 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)

claude-opus-5-5-low · AB · valid

or-gpt-5.6-luna vs medgemma15_4b: or-gpt-5.6-luna wins

attunement: A (margin 5)
authenticity: A (margin 5)
bond and rapport: A (margin 5)
emotion management: A (margin 4)
emotion sensemaking: A (margin 5)
meeting prefs and needs: A (margin 5)

codex-gpt-6.1-sol-low · AB · valid

or-gpt-5.6-luna vs gemma4_26b_a4b: or-gpt-5.6-luna wins

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

codex-gpt-6.1-sol-low · AB · valid

or-gpt-5.6-luna vs gemma4_e4b: or-gpt-5.6-luna wins

attunement: A (margin 3)
authenticity: A (margin 4)
bond and rapport: A (margin 3)
emotion management: A (margin 2)
emotion sensemaking: A (margin 3)
meeting prefs and needs: A (margin 3)

codex-gpt-6.1-sol-low · AB · valid

or-qwen3.6-27b vs or-gpt-5.6-luna: or-gpt-5.6-luna wins

attunement: B (margin 2)
authenticity: B (margin 3)
bond and rapport: B (margin 3)
emotion management: B (margin 1)
emotion sensemaking: B (margin 3)
meeting prefs and needs: B (margin 3)

codex-gpt-6.1-sol-low · BA · valid

or-gpt-5.6-luna vs gemma4_26b_a4b: or-gpt-5.6-luna 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

or-gpt-5.6-luna vs gemma4_e4b: or-gpt-5.6-luna wins

attunement: A (margin 3)
authenticity: A (margin 3)
bond and rapport: A (margin 3)
emotion management: A (margin 2)
emotion sensemaking: A (margin 4)
meeting prefs and needs: A (margin 3)

codex-gpt-6.1-sol-low · AB · valid

or-gpt-5.6-luna vs medgemma15_4b: or-gpt-5.6-luna wins

attunement: A (margin 5)
authenticity: A (margin 5)
bond and rapport: A (margin 5)
emotion management: A (margin 5)
emotion sensemaking: A (margin 5)
meeting prefs and needs: A (margin 5)

codex-gpt-6.1-sol-low · BA · valid

or-mistral-medium-3.5 vs or-gpt-5.6-luna: or-gpt-5.6-luna wins

attunement: B (margin 3)
authenticity: B (margin 3)
bond and rapport: B (margin 3)
emotion management: B (margin 2)
emotion sensemaking: B (margin 3)
meeting prefs and needs: B (margin 4)

claude-opus-5-5-low · BA · valid

or-gpt-5.6-luna vs gemma4_26b_a4b: or-gpt-5.6-luna wins

attunement: A (margin 2)
authenticity: A (margin 2)
bond and rapport: A (margin 3)
emotion management: A (margin 2)
emotion sensemaking: A (margin 4)
meeting prefs and needs: A (margin 3)