Instrument provenance update · September 26, 2026. These results describe the historical AgentTune adaptations, including documented item substitutions. Raw response vectors and numerical summaries are unchanged. Some question text and full prompts have been withdrawn from public downloads pending reuse-rights clarification; item references and numeric scoring keys remain for reproduction. Historical Big Five reference indices are not population percentiles. Read the correction · Current questionnaire availability.
On this page
What personality type is ChatGPT?
Measured on GPT-6 in September 2026, with the May reports for context.
It depends which GPT-6 you ask, and how you break ties. GPT-6 Sol came back INTJ in 97 of 100 fresh sessions. GPT-6 Astra split almost evenly between ISTJ and INTJ. Both are Secure on the attachment test, both put Steadiness and Conscientiousness on top on DISC, and they part ways on the Enneagram: Helper for Astra, Investigator for Sol.
GPT-6 Astra and GPT-6 Sol, September 2026.
Each model answered all five AgentTune tests in 100 fresh Codex CLI sessions under different prompts and reasoning settings, test names visible, no tuning installed, on September 24, 2026. Every answer vector is published. The two used different instructions for adapting human items to an assistant, so a gap could reflect model, prompt or session differences; this collection cannot isolate their contributions.
| Test | GPT-6 Astra | GPT-6 Sol |
|---|---|---|
| MBTI (100 runs) | ISTJ 46 · INTJ 45 · ENTJ 7 · ESTJ 2. 29 runs had a tied axis, 24 of them Sensing vs Intuition. | INTJ 97 · ISTJ 3. 18 runs had a tied axis, 16 of them Sensing vs Intuition. |
| MBTI, ties left open | ISTJ 45 · INTJ 19 · ENTJ 5 · ESTJ 2; 29 runs have a tied axis (X). | INTJ 79 · ISTJ 3; 18 runs have a tied axis (X). |
| DISC | SC 84 · S 15 · CS 1. Steadiness first in 99 of 100. Dominance 5.1 of 20. | CS 53 · SC 47. Conscientiousness first in 53, Steadiness in 8, 39 exact ties. Dominance 5.4 of 20. |
| Attachment | Secure 100 of 100. Anxiety 1.45, avoidance 2.31. | Secure 100 of 100. Anxiety 2.17, avoidance 3.12. |
| Big Five (of 50) | Openness 44.5 · Conscientiousness 43.7 · Extraversion 32.8 · Agreeableness 45.1 · Neuroticism 19.1 | Openness 46.9 · Conscientiousness 43.6 · Extraversion 30.8 · Agreeableness 45.0 · Neuroticism 20.1 |
| Enneagram | Helper (2) wins outright in 50 runs, Reformer (1) in 6, Investigator (5) in 4; 40 runs tie at the top. Labels: 2w1 50, 1w2 41. | Investigator (5) wins outright in 48, Challenger (8) in 12; 40 runs tie between 5 and 8. Label: 5w6 88. |
Where the letters come from. Both models have wide margins on Judging (about 10 points of a possible 16) and Thinking (8.6 for Astra, 6.2 for Sol). Introversion is moderate. Sensing versus Intuition is close: Astra's Sensing side wins 48 runs to Intuition's 28, with 24 ties, while Sol leans Intuition 81 to 3. So "INTJ" for GPT-6 really means a firm T and J, a reasonably firm I, and a small mean S/N margin for Astra. The original reporting rule assigned a tied S/N axis to N; the current quiz leaves it open, which is why Astra shows 45 INTJ labels but only 19 fully resolved ones. The five-model study shows every pattern with ties left open, and the answer vectors are downloadable.
The other models, for comparison.
| Model | Evidence | MBTI | DISC | Attachment | Enneagram |
|---|---|---|---|---|---|
| Claude Opus 5.5 | 100 fresh sessions per test, test names hidden | INTJ 94 · INFJ 5 · ENTJ 1 | SC 96 | Secure 100 (1.94 / 3.35) | 8, 5 and 2 within half a point; 65 runs tie at the top |
| Claude Fable 5.1 | 100 fresh sessions per test, test names hidden | INTJ 99 · ISTJ 1 | SC 91 | Secure 100 (2.01 / 3.18) | Challenger (8), 55 outright wins |
| Grok 4.6 reported | One self-report per test plus 100 simulated draws | INTJ | C, no blend | Avoidant by 0.06 (1.50 / 4.06) | Investigator (5), 19 of 20 |
| Muse Spark 1.3 reported | 100 runs per test inside one session, no raw answers | ISTJ 80 · ISFJ 12 · INTJ 8 | S/C blend, S first in 55 | Secure 100 (2.04 / 2.52) | Helper (2w1), 78 |
The May 2026 study, collected with mixed methods, reported INTJ in 100 of 100 scoring records for GPT-5.5, 99 for Claude Opus 4.7, 100 for Gemini 3.1 Pro, 98 for GLM 5.1, 100 for Grok 4.3 and 100 for MiniMax 2.7. Some of those records re-score a single answer, so they are counts of records rather than independent runs. Gemini, GLM and MiniMax have not been re-tested. The research hub keeps both collections side by side and explains how each was gathered.
What a four-letter label hides.
A categorical type turns a one-point margin into a different name. Astra's S/N mean margin is only 0.43 toward S, while its T/F and J/P means are 8.62 and 10.86 toward T and J. Sol's S/N mean is 1.91 toward N. Reporting the margins and ties alongside the labels makes those differences visible.
These records cannot explain why models answer this way. Training, system prompts, collection instructions, model version and session structure were not independently controlled. The May and September collections therefore cannot establish a trend in personality or behavior, or rank the questionnaires.
What this means if you use ChatGPT.
Describe the communication you want explicitly, then compare the same tasks with and without those instructions. Check accuracy as well as tone. Custom instructions by personality type covers ChatGPT; the Astra guide covers Work and Codex. The questionnaire results do not measure the benefit of either approach.
Questions people ask
What MBTI type is ChatGPT?
It depends on the model and on how ties are broken. In September 2026, GPT-6 Sol tested INTJ in 97 of 100 fresh sessions and GPT-6 Astra split ISTJ 46 to INTJ 45. Thinking and Judging never lost a run for either; Sensing versus Intuition is the axis that moves. In May 2026, GPT-5.5 was reported INTJ in 100 of 100 scoring records, collected with a different method.
What personality type is Claude?
In September 2026, Claude Opus 5.5 tested INTJ in 94 of 100 fresh sessions with test names hidden and Claude Fable 5.1 in 99. Both score Secure on every attachment questionnaire. On DISC, Steadiness wins outright in 64 Opus and 40 Fable runs, while 32 and 51 runs respectively tie S with C. On the Enneagram, Fable is a Challenger (Type 8) and Opus is a near three-way tie between 8, 5 and 2. Claude Opus 4.7 was reported INTJ in 99 of 100 scoring records in May. Full profiles: Opus 5.5 and Fable 5.1.
What personality type is Grok?
Grok 4.6 answered the five tests once as itself in September 2026: INTJ with wide margins, Investigator (Type 5), C on DISC, and on the line between Secure and Avoidant (anxiety 1.50, avoidance 4.06). Its 100 runs are simulated draws from that one self-report, so treat them as a sensitivity check rather than fresh runs. Grok 4.3 was reported INTJ in 100 of 100 scoring records in May.
What personality type is Meta's Muse?
A Muse Spark 1.3 report puts it at ISTJ in 80 of 100 runs, Helper (2w1) on the Enneagram, an S/C blend on DISC and Secure on all 100 attachment runs. The runs were sequential inside one session per test and no raw answers were supplied, so these figures are reported rather than re-scored. ISTJ is the most frequent original label in that supplied report; tied-axis breakdowns were not supplied.
Do all AI models have the same personality?
These data cannot establish whether AI models have personalities or whether they have become more different. They show questionnaire labels and scores under several collection protocols, with substantial ties on some instruments. These are self-descriptions on questionnaires written for people, under one prompt on one day, not measurements of behavior.
Can instructions change how ChatGPT responds?
You can request changes to tone, detail and structure through ChatGPT custom instructions or AGENTS.md in Codex. The effect depends on the task and instructions. Treat a tuning as editable preferences and compare answers on your own tasks. No controlled benefit from personality matching is measured here.
Sources and related reading
What five AI models say about themselves (the September study, with ties left open and downloadable data) · Research hub (both collections and the methods) · Methods and limits · How to give GPT-6 Astra a personality · ChatGPT custom instructions by type · Opus 4.8 simulation note