OpenAI Cuts GPT-5 Latency by Half After Developer Complaints

OpenAI has rolled out a performance update for its GPT-5 large-language model that roughly halves response times, according to company engineers and software developers who tested the new build on 12 August. Users of the Cursor coding environment said 95th-percentile API latency is now about twice as fast, and OpenAI’s head of developer relations noted that coding with the model "feels 2× faster." The upgrade follows criticism that GPT-5’s launch delivered slower and less reliable coding assistance than rival systems such as Anthropic’s Claude 4.1 and xAI’s Grok 4. Several developers had complained that earlier GPT-5 iterations struggled with code editing and hallucinations. Improved speed appears to be accompanied by greater stability. Tech outlet 9to5Mac reported that GPT-5 now produces more polished macOS and iOS interfaces with fewer build errors, enhancing so-called "vibe coding" workflows. The site added that Apple plans to migrate its forthcoming Siri and Apple Intelligence features from GPT-4o to GPT-5. The latest optimizations signal OpenAI’s effort to keep pace in the increasingly competitive AI-model market, where rapid iterative improvements have become critical to winning developer adoption.