OpenAI announced the release of GPT-5.5 on April 24, 2026, marking a significant update to its flagship large language model series. According to the company, the new model demonstrates superior performance in agentic coding, complex knowledge retrieval, and scientific research. Chief Executive Officer Sam Altman stated that GPT-5.5 represents a move toward more autonomous AI systems capable of executing multi-step workflows with minimal human intervention. Altman noted that the model was trained on a more diverse dataset including specialized technical journals and proprietary codebases to enhance its specialized reasoning.
A primary feature of GPT-5.5 is its computer use capability, which allows the model to interact directly with desktop environments, navigate software interfaces, and manage file systems. In internal benchmarks released by OpenAI, the model achieved a 42 percent improvement in agentic coding tasks compared to its predecessor, GPT-5. This includes the ability to debug entire software repositories and autonomously implement feature requests across multiple programming languages. The system can now interpret visual cues from a screen to operate legacy software that lacks a traditional API, a feature OpenAI refers to as Visual Action Grounding.
The model also introduces specialized optimizations for scientific research and high-level knowledge tasks. OpenAI reported that GPT-5.5 outperformed previous iterations on the Graduate-Level Google-Proof Science benchmark, showing particular strength in chemistry and physics simulations. The company noted that the model reasoning engine has been refined to handle long-context windows of up to 2 million tokens, facilitating the analysis of massive datasets and technical documentation. This allows researchers to upload entire libraries of scientific papers for cross-reference and hypothesis generation without losing context.
Alongside performance gains, OpenAI emphasized improvements in inference efficiency. GPT-5.5 is designed to operate with a 30 percent lower computational footprint than the initial release of GPT-5, significantly reducing latency for API users and lowering the cost per token for enterprise clients. To address safety concerns, the release includes an expanded suite of cybersecurity safeguards. These protocols are specifically engineered to prevent the model from assisting in the creation of malicious software or conducting unauthorized network intrusions. Chief Technology Officer Mira Murati confirmed that the model underwent six months of rigorous red-teaming by external security partners and government agencies prior to the launch.
GPT-5.5 is available starting today via OpenAI API for Tier 5 developers and through the ChatGPT Plus, Team, and Enterprise subscription tiers. The company also announced that Microsoft, its primary partner, has concurrently integrated GPT-5.5 into its Azure AI infrastructure to support global enterprise-scale deployments. OpenAI confirmed that it will continue to offer GPT-5 as a lower-cost alternative while GPT-5.5 becomes the new standard for high-complexity reasoning tasks.