ChatGPT vs. Claude personality: what default benchmarks actually show.
By Bernard Huang · Updated
Measured across 2,000 independent fresh runs on GPT-6 Astra, GPT-6 Sol, Claude Opus 5.5, and Claude Fable 5.1—plus how each responds to tuning.
On this page
September 2026 head-to-head: 400 fresh sessions per framework.
Using our open evaluation harness (reproducibility guide →), we ran each model in 100 fresh sessions per questionnaire with ties kept visible:
| Metric (n = 100 runs / model) | GPT-6 Astra | GPT-6 Sol | Claude Opus 5.5 | Claude Fable 5.1 |
|---|---|---|---|---|
| Strict resolved MBTI INTJ | 19 / 100 (29 tied; 45 ISTJ) | 79 / 100 (18 tied) | 84 / 100 (11 tied) | 98 / 100 (1 tied) |
| Big Five Agreeableness (/50) | 45.07 | 44.99 | 47.07 (highest) | 45.03 |
| Big Five Openness (/50) | 44.49 | 46.87 | 44.57 | 43.93 |
| DISC Dominance (/20) | 5.12 | 5.42 | 4.00 (floor) | 5.00 |
| DISC dominant pattern | Steadiness (99 sole S wins) | Conscientiousness (53 C, 39 ties) | Steadiness (64 S, 32 ties) | 51/100 exact S/C ties |
| Primary Customization Controls | Personalization presets + 2×1,500 chars + Projects | Personalization presets + 2×1,500 chars + Projects | Profile Prefs + Styles/Skills + Projects + CLAUDE.md | Profile Prefs + Styles/Skills + Projects + CLAUDE.md |
Notice the shared baseline: high Agreeableness and low Dominance across both labs. Claude Opus 5.5 scored highest on Agreeableness (47.07/50) and lowest on Dominance (4.00/20), while GPT-6 Astra split between ISTJ and INTJ on Sensing vs Intuition (read the tie-breaking audit →).
Where OpenAI and Anthropic differ once you add custom rules.
| Dimension | ChatGPT workflow | Claude workflow |
|---|---|---|
| Fastest global tone switch | Settings → Personalization → Base style preset + 4 toggles (guide →) | Profile Preferences box + chat Styles/Skills |
| Global instruction budget | 1,500 chars × 2 boxes + Saved Memories | Multi-paragraph text box in Profile Preferences (best kept under ~1,200 chars) |
| Workspace isolation | ChatGPT Projects (8,000 chars) or Custom GPTs | Claude Projects (system prompt + knowledge) (examples →) |
| Spoken voice persona | Voice selector inside ChatGPT Voice settings | 5 named Voice presets: Buttery, Airy, Mellow, Glassy, Rounded (guide →) |
One portable rule set that works cleanly in both ChatGPT and Claude.
Use this single behavioral block in both ChatGPT Custom Instructions (under 1,500 chars) and Claude Personal Preferences:
Test how your own prompts behave before and after in our blind Instruction A/B Test Bench, or see our three-way comparison in Muse vs ChatGPT vs Claude.
Questions people ask.
Is Claude or ChatGPT more agreeable by default?
Both score near the top of the Big Five Agreeableness scale (44.99–47.07 out of 50 in September 2026 tests), with Claude Opus 5.5 recording the highest mean Agreeableness (47.07) and lowest DISC Dominance (4.00/20).
Why do people experience GPT-6 Astra and Fable 5.1 differently if their Big Five scores look close?
Small shifts around midpoint questionnaire thresholds alter type classifications dramatically (Astra tied on 29/100 OEJTS runs while Fable resolved INTJ on 98/100), and each product pairs the base model with different default UI presets, Memory injection, and Style layers.
Can I use the same custom instructions in both ChatGPT and Claude?
Yes—if you write short behavioral rules (under 1,500 characters) instead of vague adjectives, the same instruction block fits inside ChatGPT Custom Instructions, Claude Profile Preferences, and Meta Muse Soul.md.
Does switching personality templates hurt reasoning accuracy?
Across our 60-reply pilot, explicit formatting and directness instructions lifted answer-first openings from 14/20 to 19/20 and cut hedges without lowering factual accuracy, though overly rigid length limits on complex tasks can truncate nuance.
What changed.
- First published comparing September 2026 fresh-session cohorts across GPT-6 Astra, GPT-6 Sol, Claude Opus 5.5, and Claude Fable 5.1 with cross-platform controls.