GPT-5 is here, but its unveiling lacked the sparkle many were anticipating.
The long-awaited new AI model release on Thursday was partially overshadowed by misleading charts and a false answer to a simple physics question. Also missing was the sense of transformational progress that colored past launches.
“Ok, no leap,” said François Fleuret, head of machine learning at the University of Geneva and researcher at Meta’s AI Lab, on X. ”We are seeing the plateau,” he wrote in a follow-up post. “Just scaling up is coming to an end. For everyone, not one company in particular.”
Despite a focus on its enhanced accuracy and acumen, with OpenAI CEO and co-founder Sam Altman likening the model to “talking to a PhD expert,” the model repeated a common error when explaining how wings generate lift.
But most perplexing were the series of charts the company flaunted to showcase GPT-5’s performance. Many of them had no discernible scale and misrepresented numbers. One showcased the new model’s deception rate, which visualized 47.7% as a higher bar than 50%. Another showed 52.8% as a bigger number than 69.1%. OpenAI later published corrected charts on its website, but not before the original charts started making the rounds on X.
OpenAI is under some pressure: The company is facing a talent exodus with around 25% of its research leads leaving the company over the last two years, according to a PitchBook analysis. At the same time, Altman will have to further scale up OpenAI’s commercial operation—it recently hit $12 billion in annualized revenue, per Reuters—to meet the costs of compute and to continue competing for top talent.
The first major launch since former CTO Mira Murati left to start her own company ended up as a lackluster livestream for a company that is reportedly on the verge of becoming the most valuable startup in the world.
With GPT-5, OpenAI’s coding capabilities have only now just caught up with the latest Anthropic Claude model, said Ashish Kakran, partner at Sierra Ventures. “It seems with this launch, they have closed the gap,” he said.
Enter a bright spot of GPT-5: its heavy focus on code completion and application development.
“Software on demand will be a defining characteristic of the GPT-5 era,” Altman said.
Since OpenAI’s last major product launch, the so-called “vibe-coding” industry—where AI is prompted to generate code and, in some cases, create entire applications—has exploded. AI coding startups like Lovable and Cursor are on exponential trajectories of revenue and valuation growth.
For VC investors watching on the sidelines, the coding focus was a saving grace.
“Coding is where the battleground is right now and they had to make a statement,” said Richard Dulude, co-founder and GP at pre-seed and seed investor Underscore VC. “Maybe this wasn’t as much of a quantum leap as earlier eras, but the market is increasingly competitive and keeping momentum—constant product release cadence— is the name of the game.”
https://pitchbook.com/news/articles/gpt-5s-vibe-coding-saves-the-day-in-underwhelming-openai-launch
