OpenAI has rolled out a performance update for its flagship GPT-5 model that sharply reduces response latency and boosts reasoning scores, according to developers with early access. The company introduced new caching logic and other backend changes that have cut the model’s 95th-percentile (P95) latency by about half when used inside Cursor, a popular AI-assisted code editor. Beyond speed, the latest GPT-5 build has posted gains on several proficiency tests. On internal medical-reasoning benchmarks, the system now outperforms the average human medical professional, marking a step up from the GPT-4o version that trailed clinicians on the same metrics. Independent testers also report an overall IQ estimate of 123 for the GPT-5 Pro variant. Developers say the improvements translate into a noticeably quicker coding workflow, with some describing the experience as “twice as fast.” OpenAI has not publicly detailed a release schedule for wider deployment, but the update signals the company’s continuing push to enhance both capability and efficiency ahead of a broader commercial rollout.