GPT-5.5 Landed at $5 a Million Tokens - and the Pricing Floor Just Moved Again

OpenAI shipped GPT-5.5 on April 23, 2026, and the easy headline is the benchmark sweep - strong terminal and agentic scores, the kind of numbers that get screenshotted into pitch decks. The harder, more useful headline is the price tag.

The numbers that matter

• Context window: roughly 1.05M tokens, with up to 128K tokens of output.
• API price: $5 per million input tokens, $30 per million output tokens.
• GPT-5.5 Pro: a higher-accuracy variant at $30 in / $180 out for the tasks where being right matters more than being cheap.
• Discounts: Batch and Flex tiers at half price; a Priority tier at 2.5x for latency-sensitive work.

Roll back two years and a million-token context window on a flagship model was a premium you paid through the nose for, often with a long-context surcharge bolted on top. GPT-5.5 treats it as table stakes. That's the quiet trend of 2026: the giant context window stopped being a luxury feature and became the baseline.

Why the input/output split is the whole story

Notice the gap: $5 in, $30 out. That 6x spread is not an accident, and it's the single most important thing to internalize if you're budgeting for GPT-5.5. Reading is cheap; writing is expensive. A workload that stuffs a huge document into context and asks for a short answer is wildly cheaper than one that generates long output - even if the token counts look similar at a glance.

This is exactly the kind of thing that's easy to get wrong by eyeballing. If you're sizing a GPT-5.5 workload, run your real prompts through our token calculator and split the estimate into input and output buckets before you commit to a budget. The difference between "mostly reading" and "mostly writing" can be the difference between a viable feature and a runaway bill.

Where it fits

GPT-5.5 is built for the agentic era - long tool-using loops, terminal work, code that has to actually run. The Pro variant exists for the cases where you'd rather pay 6x than ship a wrong answer: hard reasoning, high-stakes analysis, a verification pass over something cheaper.

But the launch isn't really about OpenAI flexing on a leaderboard. It's another data point in a price war that keeps dragging the floor lower. A million-token flagship at $5 input would have been unthinkable not long ago. The labs have decided that the way to win developers isn't a better demo - it's a better invoice.