OpenAI has launched GPT-5.5, which is touted as a "new level of intelligence for real work and managing agents."
Introducing GPT-5.5
— OpenAI (@OpenAI) April 23, 2026
A new class of intelligence for real work and powering agents, built to understand complex goals, use tools, check its work, and carry more tasks through to completion. It marks a new way of getting computer work done.
Now available in ChatGPT and Codex. pic.twitter.com/rPLTk99ZH5
The neural network is designed to understand complex tasks, utilize tools, verify results, and complete more tasks successfully.
The model can grasp user intentions, autonomously plan its work, and deliver results. GPT-5.5 excels at writing and debugging code, searching the internet for information, analyzing data, creating documents and spreadsheets, managing software, and switching between tools.
"Instead of meticulously controlling every step, you can assign GPT-5.5 a complex multi-step task and trust it to plan, apply tools, verify its work, navigate ambiguities, and continue working," the announcement states.
OpenAI noted that the new model is particularly effective in agent programming, computer management, intellectual work, and early scientific research—areas where establishing long chains of reasoning and actions is crucial.
"GPT-5.5 provides a leap in intelligence without sacrificing speed. Larger and more powerful models often operate slower, but GPT-5.5 matches GPT-5.4 in real-world token latency while demonstrating a significantly higher level of intelligence," the startup stated.
The neural network uses "significantly fewer" tokens when operating in Codex.
OpenAI reported implementing "the most powerful" set of safety measures before the release, collaborating with both internal and external experts.
Availability
GPT-5.5 is available in ChatGPT and Codex for users on Plus, Pro, Business, and Enterprise plans. A separate version, GPT-5.5 Pro, is available for Pro, Business, and Enterprise users.
Both variations will soon be accessible via API at a cost of $5 million for 1 million input tokens and $30 million for output tokens. The context window is 1 million tokens.
In Codex, GPT-5.5 is available for Plus, Pro, Business, Enterprise, Edu, and Go plans with a context window of 400,000 tokens. GPT-5.5 is offered in Fast mode, generating tokens 1.5 times faster at 2.5x the cost.
GPT-5.5 is more expensive than GPT-5.4, attributed to its higher token efficiency.
Capabilities of GPT-5.5
The new model consumes fewer tokens and requires fewer retries when solving tasks. In the Artificial Analysis programming index, it achieves a "leading level of intelligence" at half the cost compared to competitors.
GPT-5.5 is OpenAI's most powerful solution for agent programming. In Terminal-Bench 2.0, which tests complex command-line scenarios, it achieved an accuracy of 82.7%.
In SWE-Bench Pro, the result was 58.6%, while in Expert-SWE, the neural network outperformed GPT-5.4.
Across all three benchmarks, the new model surpassed its predecessor while using fewer tokens.
"The model's strengths in programming are particularly evident in Codex, where it can perform engineering tasks—from implementation and refactoring to debugging, testing, and validation," the company stated in its blog.
GPT-5.5 has a better understanding of system architecture: it knows why something isn't working, where corrections are needed, and which parts of the code are affected.
The model "significantly outperforms" GPT-5.4 and Claude Opus 4.7 in logical reasoning and autonomy: it proactively identifies issues, predicts testing and review needs without explicit prompts.
In the GDPval test, which assesses agents' ability to perform clearly defined intellectual tasks across 44 professions, GPT-5.5 scored 84.9%. In OSWorld-Verified, it scored 78.7%, and in Tau2-bench, it achieved 98%.
GPT-5.5 also shows strong results in other tests: 60% in FinanceAgent, 88.5% in internal modeling tasks for investment banking, and 54.1% in OfficeQA Pro.
Information Handling
GPT-5.5 is a "powerful tool for everyday computer work." The model better understands user intent and confidently navigates the entire information workflow: searching, analyzing, utilizing tools, verifying, and transforming raw data into finished results.
In Codex, GPT-5.5 outperforms GPT-5.4 in creating documents, spreadsheets, and slide presentations.
Over 85% of employees across various OpenAI departments use Codex weekly, including in software development, finance, communications, marketing, data analytics, and product management.
Scientific Research
In scientific and technical workflows, GPT-5.5 also demonstrates higher performance. This includes tasks that go beyond answering specific questions: the model can systematically explore an idea, gather evidence, test hypotheses, and interpret data.
GPT-5.5 shows improvements over GPT-5.4 on GeneBench—a platform for multi-step analysis of scientific data in genetics and quantitative biology.
In BixBench, the new model also outperformed its predecessor.
Recall that in April, OpenAI introduced "workspace agents" in ChatGPT. Teams can create shared assistants to tackle complex tasks and lengthy processes.