OpenAI released GPT-6 on September 22, 2026, saying it enables persistent agents to work for hours on complex tasks. It is the company's first model-release update since the previous GPT series.
OpenAI reported a 50% reduction in the share of prompt tokens requiring fresh processing across billions of requests, measured over the past several months. That compares with the previous baseline before the update.
GPT-6 is built on a new prompt caching system and targets applications like refactoring codebases and producing well-researched documents. Availability begins with the release of the GPT-6 family, initially for developers and businesses.
"OpenAI’s prompt caching plays a critical role in helping GitHub Copilot deliver fast, efficient experiences at scale," said Mario Rodriguez, Chief Product Officer. The result is a more efficient inference stack and faster time to first response for developers.
The announcement follows improvements in GitHub Copilot's performance. OpenAI said the update builds on prior enhancements to the inference stack, adding no judgement of your own.
OpenAI did not say how the new caching system will affect other models, and raised the question of how developers can optimize their integrations to maximize cache hit rates. The company said it will continue to refine the system as it evolves.
Source: openai
