24 July 2026 · AI & LLMs · 5 min read
Here is the part most people miss: on the API, GPT-5 is cheaper than GPT-4o. The flagship GPT-5 lists at $1.25 per 1M input tokens and $10 per 1M output — exactly half GPT-4o's input price ($2.50) at the same output price ($10), with a 400K context window instead of 128K. Newer and cheaper is not the usual direction, so it is worth doing the actual math.
Per 1M tokens, from the current OpenAI lineup:
| Model | Input /1M | Output /1M | Context |
|---|---|---|---|
| GPT-5.5 (top flagship) | $5.00 | $30.00 | 400K |
| GPT-5 (flagship) | $1.25 | $10.00 | 400K |
| GPT-5 mini | $0.40 | $1.60 | 400K |
| GPT-5 nano | $0.10 | $0.40 | 400K |
| GPT-4o (for reference) | $2.50 | $10.00 | 128K |
| GPT-4o mini (reference) | $0.15 | $0.60 | 128K |
Take a typical call: 1,000 input tokens + 500 output tokens (a short prompt with context, a medium answer). The cost is (input tokens ÷ 1M × input price) + (output tokens ÷ 1M × output price):
| Model | Cost / call | At 1M calls/month |
|---|---|---|
| GPT-5.5 | $0.0200 | $20,000 |
| GPT-4o | $0.0075 | $7,500 |
| GPT-5 | $0.00625 | $6,250 |
| GPT-5 mini | $0.0012 | $1,200 |
| GPT-5 nano | $0.0003 | $300 |
So swapping GPT-4o for GPT-5 on the same workload is about 17% cheaper — and you get 3× the context for free. Drop to GPT-5 mini and the same job costs 84% less than GPT-4o. GPT-5 nano takes it to $300/month for a million calls.
Notice that output tokens dominate every row. On the flagship, output is priced 8× the input ($10 vs $1.25). That means the cheapest way to cut a GPT-5 bill is not switching models — it is asking for shorter answers. A verbose model that writes 1,500 output tokens where 300 would do is quietly tripling the expensive half of your bill. And GPT-5.5, the premium tier at $5/$30, is 3.2× the cost of plain GPT-5 per call — worth it only for the tasks that genuinely need it.
For most production traffic the honest answer is GPT-5 mini or nano, escalating to GPT-5 only when quality demands it — the classic routing pattern. Reserve GPT-5.5 for the hardest reasoning. Run your own numbers with your real token counts before committing:
New to token pricing? Start with the free API-cost guides.
Prices are OpenAI list prices as tracked in our model catalog (July 2026) and are reference estimates — confirm current rates on OpenAI's pricing page before budgeting. Per-call figures are computed from those list prices for a 1,000-input / 500-output token call.