GPT-5.4

GPT-5.4, released on 5 March 2026, was the first OpenAI model to break past the 400,000-token ceiling the GPT-5 family had held since August 2025. Its context window is 1,050,000 tokens, output stays at 128,000, input is text and images, and the knowledge cutoff is 31 August 2025. Reasoning effort runs from none through low, medium and high to xhigh. The pricing deserves a careful read. The headline rate is $2.50 per million input tokens, $0.25 cached, and $15 per million output. But OpenAI applies a threshold: any prompt above 272,000 input tokens is billed at twice the input rate and 1.5 times the output rate — for the entire session, not merely for the tokens above the line, and in standard, batch and flex processing alike. In practice the cheap portion of the window is the first 272,000 tokens, or roughly a quarter of what the specification advertises. OpenAI released GPT-5.4 Pro alongside it for problems that reward more compute, and followed on 17 March 2026 with GPT-5.4 mini and GPT-5.4 nano for high-volume workloads. GPT-5.5 arrived seven weeks later with a full million-token window at double the price; GPT-5.4 remains in the API with no announced shutdown date.

▸Videos

No videos yet.

▸Reviews