OpenAI announces GPT-5.5 model with improved efficiency and coding

OpenAI’s ChatGPT, the San Francisco-headquartered artificial intelligence research organization, released GPT-5.5 on April 23, 2026, rolling the new model into its ChatGPT and Codex products as the company capped “a busy week of announcements” with what it described as its latest upgrade. ^[1]

OpenAI billed GPT-5.5 as its “smartest and most intuitive-to-use model yet,” framing the release as a step toward “a new way of getting work done on a computer.” ^[2] The company said the model is aimed at “agentic computer use,” including coding, web browsing and multi-step tasks that require an AI system to plan and execute actions rather than simply respond to prompts. ^[2]

Several outlets reported the launch Thursday, with TechCrunch describing GPT-5.5 as OpenAI’s newest AI model and quoting the company’s claim that it is its “smartest and most intuitive to use model” yet. ^[3] CNBC also reported that OpenAI announced the model on April 23. ^[4] The Verge likewise reported the announcement and highlighted OpenAI’s positioning of GPT-5.5 as the “next step toward a new way of getting work done on a computer.” ^[5]

A central part of OpenAI’s pitch was efficiency: the company said GPT-5.5 “match[es] GPT-5.4’s per-token latency in real-world serving” while scoring higher across benchmarks—an improvement it characterized as rare for larger models. ^[2] “Matching GPT-5.4’s per-token latency in real-world serving while hitting higher scores across benchmarks is the kind of efficiency improvement that usually doesn’t happen,” OpenAI said. ^[2]

OpenAI and outside coverage also emphasized coding and tool-using capabilities, with 9to5Mac reporting that GPT-5.5 is an upgrade for ChatGPT and Codex and that OpenAI highlighted gains in “agentic coding, computer use, knowledge work, early scientific research.” ^[1] 9to5Mac also reported that GPT-5.5 arrived seven weeks after GPT-5.4. ^[1]

The release underscores how quickly OpenAI is iterating on its flagship models, with the company and multiple outlets emphasizing both capability gains and the practical constraint that matters to many users—speed and responsiveness—through OpenAI’s claim that GPT-5.5 maintains GPT-5.4-level latency while improving benchmark performance. ^[2] OpenAI’s own framing points to where it wants the technology to go next: AI that can carry out multi-step work on a computer, particularly in coding and other “agentic” tasks, rather than only generating text. ^[2]

OpenAI’s GPT-5.5 rollout on April 23 immediately put the focus on whether the company can deliver on that “agentic computer use” promise in everyday tools like ChatGPT and Codex, as it simultaneously claims higher benchmark scores without slowing real-world serving. ^[2]