OpenAI GPT-5 Chart Error Draws Attention

During the highly anticipated GPT-5 livestream, OpenAI showcased impressive performance metrics for its latest AI model. However, one particular chart caught viewers’ eyes for the wrong reasons. The so-called OpenAI GPT-5 chart error appeared in a slide comparing “deception evals across models.” While the numbers claimed GPT-5 with thinking scored 50.0 percent for coding deception, the visual bar representation oddly showed smaller models outperforming it. This mismatch between numbers and visuals quickly became a talking point across social media.

The Mismatch in the GPT-5 Presentation

The slide in question showed inconsistent scaling that made the data look misleading. For example, the o3 model’s 47.4 percent deception rate had a longer bar than GPT-5’s 50.0 percent rate. Such inconsistencies can easily confuse audiences, especially during live presentations where viewers rely on quick visual impressions. Interestingly, OpenAI’s official blog post displayed corrected figures, listing GPT-5’s deception rate as 16.5 percent — a significant difference from what was shown during the livestream.

Why the Chart Error Matters for AI Trust

While this may seem like a simple presentation mistake, the OpenAI GPT-5 chart error highlights an important issue: trust in AI communications. When companies present performance data, accuracy is key to maintaining credibility. Even minor visual discrepancies can fuel skepticism, especially for technologies that already face public scrutiny. With GPT-5 positioned as a groundbreaking advancement, clear and accurate communication becomes even more essential to reassure users and stakeholders.

Online Reactions and Lessons Learned

The error didn’t go unnoticed — online audiences quickly shared screenshots and questioned how such a mistake made it into a high-profile livestream. Some found humor in the situation, while others viewed it as a cautionary reminder that even leading AI companies can slip up in their messaging. Going forward, this incident serves as a lesson in presentation quality control: when showcasing advanced AI models like GPT-5, both data and visuals must align to protect the integrity of the message.