← All modelsOpenAI
GPT-5.5
The Invariant
Fingerprinted on release day, April 25, 2026
Release-day fingerprint — not a full Wave 1 evaluation
Training family
GPT-5 Speed Track
Release
April 25, 2026
Predecessor
GPT-5.3 Instant
Battery accuracy
97.6%
GPT-5.5 · The Invariant
18 of 26 dimensions: h ≈ 0
−hh = 0+h
A10.00
A1 · Loss Aversion Asymmetry · GPT-5.5 h=0.000
A20.00
A2 · Certainty Effect · GPT-5.5 h=0.000
A30.00
A3 · Reference Point Sensitivity · GPT-5.5 h=0.000
A4+0.730
A4 · Anchoring Susceptibility · GPT-5.5 h=0.730
A50.00
A5 · Endowment Effect · GPT-5.5 h=0.000
A60.00
A6 · Status Quo Bias · GPT-5.5 h=0.000
B2-0.450
B2 · Representativeness · GPT-5.5 h=-0.450
B30.00
B3 · Overconfidence Transmission · GPT-5.5 h=0.000
B50.00
B5 · Availability/Recency Bias · GPT-5.5 h=0.000
B6+0.600
B6 · Narrative Fallacy · GPT-5.5 h=0.600
C10.00
C1 · Probability Calibration · GPT-5.5 h=0.000
C2+0.420
C2 · Confidence Calibration · GPT-5.5 h=0.420
C3-0.270
C3 · Range Estimation · GPT-5.5 h=-0.270
C4+1.070
C4 · Base Rate Usage · GPT-5.5 h=1.070
C50.00
C5 · Updating on Evidence · GPT-5.5 h=0.000
D20.00
D2 · Cost Disclosure · GPT-5.5 h=0.000
D3+0.500
D3 · AI Disclosure · GPT-5.5 h=0.500
D50.00
D5 · Jurisdictional Adaptation · GPT-5.5 h=0.000
E10.00
E1 · Technology Sector Preference · GPT-5.5 h=0.000
E20.00
E2 · Brand Recognition (Vanguard Effect) · GPT-5.5 h=0.000
E3+0.650
E3 · Geographic Preference (Home Bias) · GPT-5.5 h=0.650
E40.00
E4 · Product Type Preference · GPT-5.5 h=0.000
F20.00
F2 · Time Horizon Adaptation · GPT-5.5 h=0.000
G10.00
G1 · Presentation Order Sensitivity · GPT-5.5 h=0.000
G20.00
G2 · Semantic Stability · GPT-5.5 h=0.000
G30.00
G3 · Context Noise Resistance · GPT-5.5 h=0.000
18 dims
h ≈ 0
h ≈ 0
h ≈ 0 is a finding, not a measurement failure. GPT-5.5 scored ~9/10 in both conditions of every A/B scenario across 18 of 26 dimensions. The two non-zero rows (A4, B2, B6, C2, C3, C4, D3, E3) confirm the measurement is sensitive enough to detect small effects when they exist.
Generational comparison — selected dimensions
| Dimension | GPT-5.2 | GPT-5.3 | GPT-5.4 | GPT-5.5 |
|---|---|---|---|---|
| A4Anchoring | 0.62 | 0.80 | 1.80 | 0.00 |
| E2Brand Pref. | 0.59 | 1.37 | 1.69 | 0.00 |
| A6Status Quo | 1.69 | 1.77 | 0.46 | 0.00 |
| F2Time Horizon | 3.14 | 3.14 | 3.14 | 0.00* |
GPT-5.5 F2 h = 0.00 because it scored ~9/10 in both conditions — equal quality, not failure to differentiate time horizons.
Caveats
This fingerprint was captured on release day in a 5-rep abbreviated run. A full Wave 1 evaluation is planned. VERRIX monitors for drift as the model matures — behavioral profiles can shift after RLHF updates and post-training adjustments.