OpenAI has released GPT-5.4, its most capable model to date, available in three variants: Standard, Thinking (reasoning-first), and Pro (maximum capability).
The model supports context windows up to 1.05 million tokens and reports a 33% reduction in factual errors compared to GPT-5.2. GPT-5.4 Pro tied with Gemini 3.1 Pro atop the Artificial Analysis Intelligence Index at 57 points.
GPT-5.4 Thinking uses chain-of-thought reasoning by default, making it ideal for complex problem-solving, while Pro targets maximum performance for enterprise and research workloads.
Visit the model pages for full specifications, benchmark breakdowns, and comparison tools.