OpenAI's GPT-5.5 moves AI from replies to autonomous task execution
What's the story
OpenAI has unveiled its latest artificial intelligence model, GPT-5.5. The new system is said to be better at coding, using computers, and conducting deeper research than its predecessors. The announcement comes just two months after the launch of GPT-5.4, highlighting the rapid pace of innovation in the AI industry.
Enhanced features
GPT-5.5 can solve ambiguous problems with no guidance
OpenAI President Greg Brockman emphasized the unique capabilities of GPT-5.5, saying it can tackle ambiguous problems with minimal guidance. He said, "It can look at an unclear problem and figure out just what needs to happen next." This model is seen as a major step toward revolutionizing computer work and interaction.
Model proficiency
Model meets 'High' risk classification criteria
OpenAI claims that GPT-5.5 excels at data analysis, code writing and debugging, software operation, online research, document creation, and spreadsheet generation. The company also stressed that while the model meets its "High" risk classification criteria (which could amplify existing pathways to severe harm), it doesn't cross the "Critical" cybersecurity risk threshold that could open unprecedented new pathways to severe harm.
Deployment strategy
Availability and integration of GPT-5.5
OpenAI is making GPT-5.5 available to its paid subscribers, including Plus, Pro, Business, and Enterprise users in ChatGPT and Codex. The company said the model would be integrated into its application programming interface "very soon," but warned that those deployments would require "different safeguards."
Benchmark scores
Performance benchmarks and safety measures
GPT-5.5 has achieved top-tier performance on several real-world knowledge work benchmarks. It scored 84.9% on GDPval, which tests across 44 professions, and 78.7% on OSWorld-verified, which assesses various computer environments. The model also excelled in customer service workflows (98% on Tau2-bench telecom), finance, office tasks, and investment modeling. OpenAI has also strengthened safety measures for GPT-5.5 with initiatives like a biosecurity bug bounty program to test safeguards against misuse.