Model Releases
OpenAI Launches GPT-5.5: The Frontier of Agentic Intelligence
OpenAI releases GPT-5.5 on April 23, 2026, marking a historic milestone in AI with unmatched benchmark scores and token efficiency.
April 23, 2026
Model ReleaseGPT-5.5
Introduction
On April 23, 2026, OpenAI officially unveiled GPT-5.5, establishing it as the smartest and most intuitive language model in the company's history. This release represents a significant leap forward in agentic capabilities, designed not just to answer questions, but to execute complex workflows on a computer autonomously. As a frontier model with undisclosed parameters, GPT-5.5 signals a new era of human-AI collaboration where the machine handles the heavy lifting of reasoning and execution.
The model is positioned as the next step toward a fundamentally new way of getting work done on a computer. Unlike previous iterations, GPT-5.5 focuses heavily on reducing latency while increasing intelligence, ensuring that developers and engineers experience a seamless interaction without sacrificing performance. This rollout marks a critical turning point for the industry, outperforming key competitors like Claude Opus 4.7 and Gemini 3.1 Pro across most professional benchmarks.
- Released: April 23, 2026
- Status: Frontier Model (Undisclosed Parameters)
- Provider: OpenAI
Key Features & Architecture
GPT-5.5 is co-designed, trained with, and served on NVIDIA GB200 and GB300 NVL72 systems, leveraging cutting-edge hardware to achieve unprecedented efficiency. The architecture prioritizes token efficiency, allowing the model to complete tasks using fewer tokens compared to GPT-5.4 while maintaining or improving per-token latency. This efficiency translates to faster response times and lower operational costs for enterprise users.
Security is paramount in this release. The model is proactively deployed with industry-leading cybersecurity safeguards and is classified as High under the OpenAI Preparedness Framework. Additionally, GPT-5.5 Thinking unlocks faster help for harder problems by providing smarter, more concise answers, reducing the cognitive load on the user while maximizing output quality.
- Hardware: NVIDIA GB200 and GB300 NVL72 systems
- Efficiency: Significantly more token efficient than GPT-5.4
- Security: High classification under OpenAI Preparedness Framework
Performance & Benchmarks
The performance metrics for GPT-5.5 are state-of-the-art across a wide array of technical challenges. On Terminal-Bench 2.0, the model achieves an impressive 82.7%, demonstrating superior agentic coding capabilities. In the realm of software engineering, it scores 58.6% on SWE-Bench Pro, solving more tasks end-to-end in a single pass than previous models. Furthermore, it reaches 78.7% on OSWorld-Verified, confirming its ability to operate effectively within real computer environments.