OpenAI has rolled out server-side changes to GPT-5 that markedly speed up the model’s responses in Cursor, a popular AI-assisted code editor. Developers testing the updated interface report that 95th-percentile latency is now roughly twice as fast after improvements to caching and API throughput. The upgrade arrives alongside a pricing revision that reduces the cost of cached or repetitive input tokens to one-tenth of the standard rate, down from one-quarter previously. Users say the lower fee makes long programming sessions and automated refactoring significantly cheaper. While the base GPT-5 model now executes faster, developers note that capability varies by subscription tier. ChatGPT Plus and Team accounts provide a smaller context window and lower "GPT-5 Thinking" setting than Pro accounts, which continue to offer the higher-reasoning configuration used in Cursor’s “gpt-5-high” option.