GPT-5.5 (medium) is a specialized configuration of OpenAI's frontier language model series, optimized for complex, real-world agentic workflows. Released on April 23, 2026, it represents a specific reasoning effort level within the GPT-5.5 family designed to balance high-level cognitive performance with token efficiency. The model is engineered to function as a "new class of intelligence," capable of executing multi-step tasks such as repository-scale coding, autonomous online research, and cross-tool document management with minimal human intervention.
A defining characteristic of GPT-5.5 (medium) is its reasoning-centric architecture, which utilizes reinforcement learning to produce a long internal chain of thought before responding. This allows the model to refine its strategies and self-correct during the execution of difficult tasks. Despite its increased capability, it matches the per-token latency of previous generations while consuming approximately 40% fewer tokens for the same output, making it significantly more efficient for high-volume professional workloads.
The model's performance is particularly notable in agentic coding and computer use, where it has set benchmarks on evaluations like Terminal-Bench 2.0 (scoring 82.7%) and OSWorld-Verified. These capabilities allow it to operate software independently, diagnose architectural-level failures in codebases, and manage complex spreadsheets or presentation decks. However, third-party evaluations suggest that while the model leads in raw reasoning and accuracy, it may exhibit higher rates of confident hallucination in factual recall compared to its more conservative predecessors.
Technically, GPT-5.5 (medium) supports an extensive 1 million token context window in its API implementation, facilitating the analysis of vast datasets and long-form technical documentation. While OpenAI has not publicly disclosed the exact parameter count, the model was trained using massive compute resources, including the Stargate facility. It is primarily accessed through the OpenAI API and flagship subscription services, serving as a foundational engine for autonomous AI agents in enterprise and research environments.