Migrating: GPT-5.4 Pro to GPT-5.5 Pro

A premium-to-premium move where Standard pricing doesn't move at all — every input and output rate matches between these two, at every tier both publish.

What actually moves

The knowledge cutoff advances to GPT-5.5 Pro's newer training data, which with prices unchanged is the main benefit. The default reasoning effort moves from medium to high — the same three levels are accepted, but an unspecified request will reason more and spend more reasoning tokens after the switch. Set effort explicitly if you want the old behavior.

Rate limits are the part to re-check row by row. On the Build, Launch and Grow ladder the two allow the same requests per minute at Build and Launch, and GPT-5.5 Pro allows more at Grow. GPT-5.5 Pro offers larger token ceilings at Build and Launch, while GPT-5.4 Pro keeps the larger batch queue at every tier, so a batch-heavy workflow can lose queue headroom by moving. GPT-5.4 Pro also publishes a separate long-context table for prompts over 272,000 input tokens; GPT-5.5 Pro's page publishes none, so check your organization's limits page for how it treats large prompts.

What stays put

Context window, output ceiling, the absence of Fast mode and of a cached-input price. GPT-5.4 Pro publishes long-context bands for Standard, Batch and Flex; GPT-5.5 Pro's page doesn't state the rule, and only a Standard long band appears in its pricing row — so a large-prompt batch job loses a published long-context Batch price in this move.

Before switching

Run the real request shape through the rate-limit tier planner against both models at your actual tier. And consider whether a Pro model is needed at all: reasoning.mode: "pro" on GPT-5.6 and GPT-6 performs extra model work billed at that model's own standard rates.