Did OpenAI show misleading GPT-5 charts during launch?
What's the story
OpenAI's recent launch of its latest AI model, GPT-5, was marred by some glaring errors in the charts presented during the live-stream. The company had showcased several comparisons to highlight the new model's capabilities over its predecessors and competitors. However, some of these charts were found to be misleading, prompting an apology from an OpenAI employee.
Chart inaccuracies
One chart showed wrong data for GPT-5
One of the misleading charts compared GPT-5 with OpenAI's o3 on various parameters. The parameter 'coding deception' showed a 50% deception rate for GPT-5 against o3's score of 47.4%, a statistically close call, but the graph told a different story. Instead of showing a higher bar for GPT-5, the chart depicted o3 with a much bigger bar. OpenAI later issued a correction in a blog post, clarifying that the actual deception rate for GPT-5 is 16.5%.
Additional errors
Another downplayed difference between GPT-5 and other models
Another misleading chart compared GPT-5 with o3 and GPT-4o, showing a score of 74.9 for the new model against 69.1 and 30.8 for the other two models, respectively. Here too, the difference in scores was downplayed as o3 and GPT-4o were shown with almost equal length bars while GPT-5 had a bigger bar. OpenAI CEO Sam Altman termed this incident a "mega chart screwup" and confirmed that an accurate version of these charts has been uploaded on OpenAI's blog post.
Twitter Post
'Worst chart crimes of the century'
this screenshot from GPT-5 livestream has to be among the worst chart crimes of the century pic.twitter.com/HXsK2CWCon
— Ege Erdil (@EgeErdil2) August 7, 2025
Information
Incident raises questions about how tech companies present data
An OpenAI marketing team member also acknowledged the mistake in a post on X, saying, "We fixed the chart in the blog guys, apologies for the unintentional chart crime." The incident has sparked conversations about how tech companies should be more cautious with data visuals.
Twitter Post
OpenAI member acknowledging the mistake
We fixed the chart in the blog guys, apologies for the unintentional chart crime 🙏
— pranav (@pranaveight) August 7, 2025
Can't wait for you to start using GPT-5 as we start rolling it out today!https://t.co/1x6yZ4k7TUpic.twitter.com/M1IchNbHwd
Model advancements
Everything to know about GPT-5
Despite the chart mishaps, OpenAI has touted GPT-5 as a major upgrade over its predecessor, the company's GPT-4o model. The new model is said to be more accurate, faster, and better at reasoning, context recognition, structured thinking, and problem-solving. A key innovation with GPT-5 is a unified system that employs an 'efficient model' for regular tasks and a dedicated reasoning model called "GPT-5 Thinking" for more complex reasoning-based tasks.