OpenAI releases GPT-5.5 and brings ChatGPT closer to AI super-app format

OpenAI has released GPT-5.5 and is positioning it not as another incremental ChatGPT update, but as part of a larger strategy: to transform its products into a unified work interface where the model not only answers questions but also independently works through task steps, uses tools, and brings work to completion. For the market, this is an important signal: the AI assistant race is shifting from the quality of individual responses to the quality of autonomous task execution on a computer.

The announcement took place on April 23, 2026. The company calls GPT-5.5 its smartest and most intuitive model release to date.

According to OpenAI president Greg Brockman, the new version brings the company closer to a more agentic and natural way of working with computers. This is not just about a chatbot, but a system you can give a vague multi-step task to — write and debug code, gather materials online, analyze data, format a document or spreadsheet — and get not just text, but a completed result. According to OpenAI's official materials, GPT-5.

5 maintains latency at the GPT-5.4 level but solves the same tasks with fewer tokens and performs better on specialized tests. On Terminal-Bench 2.

0, which tests complex command-line work, the model scored 82.7% versus 75.1% for GPT-5.

4. On OSWorld-Verified, which evaluates independent computer environment work, — 78.7%.

On GDPval, which measures the quality of formalized intellectual work across dozens of professions, — 84.9%. OpenAI also claims that GPT-5.

5 outperforms previous models and in several scenarios beats strong competitors like Claude Opus 4.7 and Gemini 3.1 Pro.

The main emphasis is placed on two classes of tasks: agent development and intellectual office work. In Codex, the model should better maintain context in large projects, identify causes of failures, plan fixes, and verify its own work with tools. In regular work scenarios — gathering information, building tables and presentations, assisting with analytics, finance, and research tasks.

OpenAI provides internal usage examples: over 85% of company employees already use Codex on a weekly basis, and individual teams are automating reports, request processing, and extensive document work. This logic is directly tied to the AI super-app idea: in a corporate note from April 8, OpenAI described a unified interface that would combine ChatGPT, Codex, agentic web browsing, and actions in other tools. Another important area is scientific and technical research.

The company states that GPT-5.5 has significantly strengthened in tasks where you need to not just provide an answer, but complete a full cycle: gather data, test hypotheses, interpret results, and suggest the next step. Examples include improvements on GeneBench and BixBench, as well as cases where the model helped with biological data analysis and mathematical proofs.

At the same time, OpenAI has strengthened restrictions around cybersecurity and biological risks: the model underwent additional testing, and for sensitive scenarios the company introduced stricter classifiers and access modes. On April 23, GPT-5.5 began rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, and on April 24, OpenAI clarified that GPT-5.

5 and GPT-5.5 Pro became available via API as well. What does this mean?

OpenAI increasingly looks less like a single chatbot developer and more like a provider of a universal work environment where the model takes on not just text generation but also part of execution. If GPT-5.5 truly works consistently in long chains of actions, the market will get a new standard: models will be compared not by beautiful demos and single responses, but by how reliably they can bring complex tasks to completion without constant manual oversight.