“We totally screwed up some things”: Sam Altman on GPT-5 backlash and OpenAI’s lessons ahead of GPT-6 launch

OpenAI CEO Sam Altman

OpenAI CEO Sam Altman has admitted that the launch of GPT-5 did not meet expectations, describing it as a “rocky rollout” that taught the company several valuable lessons. Altman said the upcoming GPT-6 will build on those insights, addressing user feedback and performance inconsistencies that surfaced after GPT-5’s debut earlier this year.

The GPT-5 backlash

GPT-5, introduced in August 2025, was expected to mark a major leap in OpenAI’s large language model series. However, many users said the improvements felt incremental. Some described the model as less creative and more mechanical than GPT-4o, while others reported inconsistent reasoning and slower response times.

Altman later acknowledged that internal system changes had created unforeseen issues. “We totally screwed up some things on the rollout,” he said in a recent interview. He explained that a routing mechanism designed to delegate tasks to smaller sub-models sometimes misfired, making GPT-5 appear less capable than intended.

Learning from mistakes

Altman said the company underestimated how much users valued GPT-4o’s conversational personality. “We for sure missed how much people cared about the warmth and tone,” he said, adding that technical progress alone cannot define user satisfaction. OpenAI responded by reinstating GPT-4o for paid users, expanding access to reasoning tools, and improving transparency by showing which model was active in ChatGPT sessions.

Looking ahead to GPT-6

Altman said GPT-6 will be a “significant upgrade” in both capability and design. The new model is expected to offer better contextual understanding, more consistent reasoning, and improved user adaptability. “GPT-6 will reflect everything we’ve learned from GPT-5,” Altman said. “The vibes were kind of bad at launch, but now they’re great.”

He also hinted that GPT-5 has already begun contributing to research and education in new ways, but GPT-6 will take those applications further as OpenAI aims for more reliable, human-aligned AI systems.