OpenAI's GPT-5.5: A Revolution in Fast and Efficient AI

GPT-5.5: A Major Advancement in Artificial Intelligence

OpenAI has unveiled GPT-5.5, an artificial intelligence model that marks a new milestone in how tasks are accomplished on a computer. This model stands out for its ability to quickly understand user intentions and autonomously manage a significant portion of the work.

GPT-5.5 excels in various areas such as writing and debugging code, online research, data analysis, as well as document and spreadsheet creation. It can navigate between different tools until a task is completed, allowing users to delegate complex tasks without having to manage each step individually. The gains are particularly notable in agentic coding, computer usage, knowledge work, and scientific research, where progress relies on contextual reasoning and action over time.

GPT-5.5 achieves this advancement in intelligence without sacrificing speed. While larger and more powerful models are often slower, GPT-5.5 maintains the same latency per token as GPT-5.4 in real-world conditions, while offering a higher level of intelligence. It also uses fewer tokens to accomplish the same Codex tasks, making it both more efficient and effective.

OpenAI launched GPT-5.5 with a set of enhanced safeguards to reduce abuse while preserving access for beneficial purposes. The model has been evaluated through a comprehensive suite of safety and readiness frameworks, in collaboration with both internal and external teams. Targeted testing has been added for advanced capabilities in cybersecurity and biology, and feedback has been gathered from nearly 200 trusted partners prior to launch.

Today, GPT-5.5 is available for Plus, Pro, Business, and Enterprise users in ChatGPT and Codex. The GPT-5.5 Pro model is also deployed for Pro, Business, and Enterprise users in ChatGPT. API deployments require different safeguards, and OpenAI is working closely with its partners and clients on security requirements for large-scale deployment. GPT-5.5 and GPT-5.5 Pro will soon be available via the API.

Model Capabilities

OpenAI is building a global infrastructure for agentic AI, enabling individuals and businesses around the world to accomplish their work with the help of AI. Over the past year, a significant acceleration in software engineering through AI has been observed. With GPT-5.5 in Codex and ChatGPT, this transformation is beginning to extend to scientific research and the broader work done on computers.

In these areas, GPT-5.5 is not only smarter; it is also more efficient in its problem-solving approach, often achieving better quality results with fewer tokens and fewer attempts. On the Artificial Intelligence Coding Index, GPT-5.5 offers cutting-edge intelligence at half the price compared to competing coding models.

Agentic Coding

GPT-5.5 is OpenAI's most powerful agentic coding model to date. On Terminal-Bench 2.0, which tests complex command-line workflows requiring planning, iteration, and tool coordination, it achieves a peak accuracy of 82.7%. On SWE-Bench Pro, which evaluates real-world GitHub problem-solving, it reaches 58.6%, solving more end-to-end tasks in a single pass than previous models. On Expert-SWE, our internal assessment for long-term coding tasks with a median estimated human completion time of 20 hours, GPT-5.5 also outperforms GPT-5.4.

The model's coding strengths are particularly evident in Codex, where it can support engineering work ranging from implementation and refactoring to debugging, testing, and validation. Preliminary tests suggest that GPT-5.5 excels in behaviors critical to real engineering work, such as maintaining context across large systems, reasoning through ambiguous failures, verifying hypotheses with tools, and making changes across the entire codebase.

Knowledge Work

The same strengths that make GPT-5.5 excellent for coding also make it powerful for everyday work on a computer. With a better understanding of intent, the model can navigate more naturally through the entire knowledge work cycle: finding information, understanding what is important, using tools, verifying results, and transforming raw materials into something useful.

In Codex, GPT-5.5 outperforms GPT-5.4 in generating documents, spreadsheets, and presentations. Alpha testers reported that it surpassed previous models in tasks such as operational research, spreadsheet modeling, and transforming messy business inputs into plans. When combined with Codex's computer usage skills, GPT-5.5 brings us closer to the sensation that the model can actually use the computer with you: seeing what is on the screen, clicking, typing, navigating interfaces, and moving between tools with precision.

OpenAI teams are already leveraging these strengths in real workflows. Today, over 85% of the company uses Codex weekly in functions such as software engineering, finance, communications, marketing, data science, and product management.

In ChatGPT, GPT-5.5 Thinking unlocks faster assistance for more challenging problems, with smarter and more concise responses to help you progress through complex work more effectively. It excels in professional tasks such as coding, research, synthesis and analysis of information, and document-heavy tasks, especially when using plugins.

In GPT-5.5 Pro, early testers have noted a significant advancement in both the difficulty and quality of work that ChatGPT can accomplish, with latency improvements making it much more practical for demanding tasks. Compared to GPT-5.4 Pro, testers found that GPT-5.5 Pro's responses were significantly more complete, well-structured, accurate, relevant, and useful, with particularly strong performance in the fields of business, law, education, and data science.