OpenAI has introduced a new AI model, GPT-5.5, quietly pushing its technology a step closer to working more independently.
The idea is simple: instead of guiding the system step by step, users can now hand over a task and expect it to figure things out on its own.
The model is rolling out to paid ChatGPT users, including those on Plus, Pro, Business, and Enterprise plans. It’s also being added to Codex, with API access expected soon for developers.
What stands out in GPT-5.5 is how it handles instructions. Rather than needing carefully structured prompts, it can take in messy or multi-part requests and turn them into a finished result. The system is built to plan its approach, use tools where needed, check its own work, and refine outputs along the way.
This marks a noticeable change in direction. OpenAI’s earlier models focused on responding to prompts. GPT-5.5 leans more toward actually doing the work—whether that’s writing code, analyzing data, preparing documents, or handling research tasks.
Early test results suggest the improvements are real. The model scored 82.7% on Terminal-Bench 2.0, which looks at complex command-line tasks, and 78.7% on OSWorld-Verified, a benchmark that measures how well AI operates in real computing environments. On SWE-Bench Pro, a test focused on software debugging, it reached 58.6%, showing stronger performance in fixing issues in one go.
The gains aren’t limited to coding. On GDPval, which measures performance across dozens of professional tasks, GPT-5.5 scored 84.9%. That points to better results in areas like research, writing, and data analysis. It has also shown improvement in scientific workflows, including bioinformatics and multi-step data processing.
Another important change is efficiency. OpenAI says the model delivers similar speed to its predecessor but uses fewer tokens overall. In simple terms, it can get more done with less back-and-forth, which may help reduce costs and improve usability.
Inside Codex, GPT-5.5 is designed to handle entire software projects—from writing and debugging code to testing and refining it—while keeping track of larger codebases. Early users have pointed to clearer reasoning and a better ability to understand problems before solving them.
Pricing for the API starts at $5 per million input tokens and $30 per million output tokens, with higher tiers for advanced versions. The company expects that improved efficiency will balance out the higher capability.
Alongside these upgrades, OpenAI is also tightening safety measures. The model has been tested for risks related to cybersecurity and sensitive scientific areas, with additional safeguards added for higher-risk use cases.
The company is also expanding its Trusted Access for Cyber program, giving verified organizations broader access to advanced tools for defensive security work. This comes shortly after the release of GPT-5.4-Cyber, a model focused specifically on cybersecurity.
Overall, GPT-5.5 feels less like a tool that waits for instructions and more like one that can take ownership of a task and carry it through. That shift, even more than the benchmarks, may define how people end up using it day to day.
Also Read: Cognizant, OpenAI Expand Codex Across Enterprise Software Engineering
