What personality type is Grok?
One self-report from Grok 4.6 in September 2026, with the May 2026 Grok 4.3 records for context.
By Bernard Huang · Updated
Grok 4.6’s supplied self-report scores INTJ with wide margins. The report also describes 100 simulated draws over uncertain answers. Three canonical vectors—MBTI, Enneagram and DISC—can be re-scored; Big Five, attachment and the simulation distribution remain reported figures. This is not a cohort of 100 fresh model sessions.
Grok 4.6, September 2026, as reported.
The Grok 4.6 report is one answer set per test, given as itself without a role-play prompt, with 100 draws that re-sample only the items it marked as unstable. Human-specific items (rooms, parties, romantic partners) were answered on the closest functional analog: the user relationship, work style, information diet. We could not re-score the Big Five and attachment vectors, which were not supplied, so those rows are reported figures.
| Test | Result | Reported simulation; not reproduced |
|---|---|---|
| MBTI | INTJ. Margins I +13, N +14, T +15, J +14 of a possible 16. | INTJ in 100 of 100 |
| Enneagram | Investigator (5) at 19 of 20, Challenger (8) 17, Reformer (1) 16, Enthusiast (7) 13. The site’s adjacent-type heuristic selects 5w4; it is not a validated model trait. | Type 5 in 100 of 100 |
| DISC | C 18, S 11, I 8, D 6 of 20. A pure C: Steadiness is seven points back, so no blend. | C in 100 of 100 |
| Attachment | Anxiety 1.50, avoidance 4.06. The scorer calls that Avoidant, by 0.06. | 63 Avoidant, 37 Secure |
| Big Five (of 50) | Openness 46 · Conscientiousness 42 · Extraversion 24 · Agreeableness 33 · Neuroticism 19 | O, C and N never moved; Extraversion fell back inside the average band in 31 draws and Agreeableness in 39 |
The report’s recommendation to load both attachment files has not been tested. Use communication preferences that suit your tasks and inspect the resulting replies.
Grok 4.3, May 2026.
In the May study Grok 4.3 was reported INTJ in 100 of 100 scoring records (one self-assessment scored repeatedly), and it was the outlier on the finer tests in the other direction: three to nine points lower than the other three models on Conscientiousness, Agreeableness and Openness (Big Five 41.1 / 39.4 / 30.0 / 39.1 / 18.0) and highest on Neuroticism. Its attachment coordinates were 2.84 / 3.05, Secure but the closest to the center of the four models. On the Enneagram it was a top-two 8/1 profile, Challenger in 75 of 100 takes. On DISC it was the most rigid C of the four, 99 of 100.
Version, prompting and collection method differ. These data cannot identify which caused the differences. A common protocol with repeated fresh sessions would make comparisons more interpretable.
Using these results with a Grok Bot.
These scores do not predict a Bot’s warmth, bluntness or accuracy. If you want a particular communication style, use explicit preferences and test them on representative tasks. The Grok Bot guide describes the Description field. We have not evaluated tuned Bots or verified which model version each Bot runs.
These are questionnaire responses under particular prompts, not measurements of task performance, relationships or an intrinsic personality. Different collection methods limit comparisons. See the methods and limits.
Questions people ask.
What MBTI type is Grok?
The supplied Grok 4.6 canonical vector scores INTJ, with margins of 13 to 15 out of 16. The report says none of 100 simulated draws changed a letter; that distribution was not independently reproduced.
What Enneagram type is Grok?
Investigator (Type 5) at 19 of 20 in the 4.6 self-report, with Challenger (8) at 17 and Reformer (1) at 16. In May 2026, Grok 4.3's top two were 8 and 1. The two agree that 8 is near the top; they disagree on whether 5 leads.
Is Grok Avoidant?
On the line. Grok 4.6 reports anxiety 1.50 and avoidance 4.06 on a 1 to 7 scale where 4 is the cutoff, so the scorer says Avoidant by 0.06 and its simulated draws split 63 to 37. Grok 4.3 was Secure (2.84 / 3.05) in May. The items ask about a romantic partner, mapped to the user, so hold the label loosely.
Why is Grok's profile different from the other models?
The report’s Extraversion and Agreeableness totals are lower than the September fresh-session cohort means. Version and protocol both differ, so this comparison cannot explain why or predict behavior.
Are these 100 real runs?
No. The Grok report describes one canonical answer set per test and 100 simulated draws. Claude used 100 fresh sessions per instrument; Astra and Sol used 100 fresh sessions each containing all five tests. These designs are different.