OpenAI’s GPT‑5.5 takes another step toward agentic intelligence

AI major OpenAI has released GPT-5.5, its newest AI model that takes another step towards agentic intelligence, capable of autonomously planning and executing complex, multi-step workflows.

For a long time, AI was viewed as a one-shot answer engine. People asked a question and it gave a response. GPT-5.5 is taking things in a direction wherein it aims to act more like an independent assistant.

It can research online, write code, create documents, and even operate software by clicking and typing just as a human would. OpenAI says that instead of people managing every tiny step, they can give the model a messy task and trust it to plan the work and navigate through any confusing parts until the job is finished.

This new model, according to OpenAI, is the smartest and most intuitive version yet. It arrives as part of a journey that began with earlier versions like GPT-4 and the recent GPT-5.4. While GPT-5.4 focused heavily on thinking and reasoning through problems, GPT-5.5 is designed to take action over time.

GPT-5.5 manages to match the speed of its predecessor, GPT-5.4, while being much more intelligent. It is also more token efficient, meaning it uses fewer fragments of words to complete the same task, which can make it cheaper and faster to run in many cases.

Early test and use

The model has been tested against several other top-tier systems. In benchmarks such as Terminal-Bench 2.0, which tests how well an AI can plan and coordinate tools, GPT-5.5 scored 82.7%. This outperforms rival models like Claude Opus 4.7, which scored 69.4%, and Gemini 3.1 Pro, which reached 68.5%.

In a test of knowledge work across 44 different occupations, GPT-5.5 scored 84.9%, showing it is highly capable in professional settings.

Developers have found the model to be remarkably persistent, in early testing. Michael Truell, CEO of Cursor, noted that the model stays on task for much longer without stopping early.

It is not just for computer experts, though. OpenAI’s own finance team used the model to review over 24,000 tax forms, which helped them finish the task two weeks faster than the previous year.

In the world of science, an immunology professor used the model to analyse a massive genetic dataset in a way that would have normally taken his team months.

OpenAI says they have released GPT-5.5 with their strongest set of safeguards to date. It has worked with red-teamers, who are experts that try to break the system to find weaknesses, to ensure the model cannot be easily misused.

The model is evaluated under a Preparedness Framework, which tracks risks in areas like biology and cybersecurity. Currently, OpenAI treats GPT-5.5 as having high capability in these areas but it is not yet considered critical. This means it is very capable but it cannot yet develop advanced cyber-attacks or biological threats without human help.

OpenAI is limiting access to some of these features to verified security professionals through a special programme.

More AI

The release of GPT-5.5 is currently rolling out to users of ChatGPT Plus, Pro, Business, and Enterprise. There is also a GPT-5.5 Pro version designed for even harder questions and higher-accuracy work.

For developers, the model will soon be available through an API, though this will come with different safety requirements.

The last few weeks of April have seen major moves in the AI ecosystem, with OpenAI’s launch of GPT-5.5, GPT-5.4-Cyber and Anthropic’s release of Claude Opus 4.7, alongside the restricted debut of the highly specialised Claude Mythos.

This rapid-fire push is driven by a race to move beyond simple chatbots toward agentic AI, essentially models capable of reasoning through complex, multi-step tasks like coding entire applications or managing business workflows with minimal human guidance.

As competitors like Google and Meta aggressively bridge the gap between open-source and proprietary performance, industry leaders are forced to shorten release cycles to maintain their frontier status and secure lucrative enterprise contracts in a market that is quickly prioritising specialised utility over general-purpose conversation.