GPT-5.5 System Card Figures Inconsistent

In OpenAI's GPT-5.5 system card, two figures that appear to present the same evaluation results reached different conclusions.

• In Figure 1(p.6), GPT-5.5 appears to produce more overconfident answers and fabricated facts than GPT-5.4.
• In Figure 4(p.13), the same items are shown at lower rates instead, reading as though GPT-5.5 is better.
• The body text also swings between the two extremes. The system card states that GPT-5.5 shows a mix of higher and lower rates of misalignment compared to GPT-5.4 across various items.

In other words, the document contains conflicting figures and interpretations side by side, revealing the possibility that the evaluation process or document management was not properly organized.