According to OpenAI, GPT-5.5 is its most capable agentic coding model so far. It can quickly understand user intent and execute multi-step tasks with minimal guidance. The model is built to perform a wide range of functions, including writing and debugging code, conducting research, analyzing data, generating documents and spreadsheets, and even navigating software tools to complete tasks end-to-end.
Speaking during a media briefing on Thursday (April 23), OpenAI President Greg Brockman said the model’s defining strength lies in its ability to work through ambiguous problems on its own. He noted that it can “look at an unclear problem and figure out what needs to happen next,” adding that it signals a shift in how people may use computers in the future.
Stronger coding and automation capabilities
One of the biggest improvements is in coding performance. GPT-5.5 can handle complex programming workflows, fix bugs, test code, and manage large codebases with greater accuracy. Company benchmarks indicate notable gains over its predecessor, GPT-5.4, particularly in solving real-world coding problems in a single attempt.
Early testers, including engineers, reported improved reasoning abilities, with the model able to anticipate issues and identify testing or review requirements without explicit prompts.
The model is also more efficient, producing higher-quality outputs while using fewer tokens for tasks. Despite these improvements, OpenAI says GPT-5.5 maintains response speeds similar to earlier versions, addressing the usual trade-off between performance and latency.
Productivity and “computer use” features
Beyond coding, GPT-5.5 is positioned as a tool for everyday knowledge work. It can assist with research, generate structured reports, build spreadsheets, and summarize complex information. The model is also designed to interact more naturally with software, helping users navigate interfaces, manage workflows, and move across tools more seamlessly.
OpenAI said its Codex system is already being used internally across departments such as engineering, finance, and communications to automate workflows. More than 85% of employees reportedly use Codex weekly across functions including marketing, data science, and product management.
safety measures
The company said GPT-5.5 has been released with enhanced safety systems. These include preparedness evaluations, domain-specific testing, targeted checks for advanced biology and cybersecurity capabilities, and assessments conducted with external experts.
Availability and pricing
GPT-5.5 is being rolled out to Plus, Pro, Business, and Enterprise users across ChatGPT and Codex. The Pro version is currently available for Business and Enterprise tiers, with broader API access expected soon.
For developers, the standard model is priced at $5 per million input tokens and $30 per million output tokens. The higher-accuracy Pro version is priced at $30 and $180, respectively. OpenAI is also offering Batch and Flex pricing options at a 50% discount, along with premium Priority processing.
Despite being more expensive than GPT-5.4, the company said GPT-5.5 is designed to be more efficient overall, delivering better results while using fewer tokens.
