GPT-5.5 API pricing, explained
What GPT-5.5 costs per million tokens, a worked monthly example, and when a smaller GPT model is enough.
The price per million tokens
Like every large language model API, GPT-5.5 is billed per token, in two rates: one for the tokens you send (input) and one for the tokens it writes back (output). A token is roughly three quarters of an English word. The table shows GPT-5.5 next to the GPT models people usually weigh it against.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| GPT-5.5 openai/gpt-5.5 | ₹494 | ₹2,962 | 1.05M |
| GPT-5.4 openai/gpt-5.4 | ₹247 | ₹1,481 | 1.05M |
| GPT-5.4 mini openai/gpt-5.4-mini | ₹74.05 | ₹444 | 400K |
| GPT-5.4 nano openai/gpt-5.4-nano | ₹19.75 | ₹123 | 400K |
These are OpenAI's list prices, which is exactly what Rapid Router debits per request. Our platform fee is charged once, when you buy credits, so nothing is added on top of each call.
A worked monthly example
Take an assistant that reads 10 million input tokens and writes 2 million output tokens in a month: a few thousand conversations with a decent system prompt. On GPT-5.5 that is ₹10,861. The same month on GPT-5.4 mini is ₹1,629.
The cost calculator runs the same sum for your own volumes across every live model.
Where the bill really comes from
Three things decide most of a GPT-5.5 bill. First, output: at several times the input rate, long answers cost more than long prompts. Second, reasoning: GPT-5.5 thinks before it answers and those tokens are billed as output. Third, repetition: if every request resends the same long system prompt or document, you pay to read it again each time unless the provider caches it.
Rapid Router passes prompt caching through automatically: when OpenAI reports cached tokens, they are billed at the cache rate, and each response shows the saving. Keep the fixed part of your prompt first and the changing part last to get hits. More in the prompt caching docs.
When to use a smaller model
GPT-5.5 earns its price on hard reasoning, long multi-step work and code. For extraction, classification, summaries and routine replies, a smaller model usually gives the same answer for far less. A common setup sends everything to the small model and escalates only the requests it flags as hard.
The best cheap AI model ranking and the GPT-5.5 model page help you choose. To call GPT-5.5, point the OpenAI SDK at https://rapidrouter.ai/api/v1 and set the model to openai/gpt-5.5.
Common questions
How much does the GPT-5.5 API cost?
On Rapid Router, GPT-5.5 is billed at OpenAI's list price: ₹494 per million input tokens and ₹2,962 per million output tokens, with a platform fee charged once when you buy credits.
Why is output so much more expensive than input?
Generating a token takes far more compute than reading one, so every provider prices output several times higher. For GPT-5.5 the gap is the main driver of your bill: a chatty answer costs more than a long prompt.
Do reasoning tokens count as output?
Yes. GPT-5.5 thinks before it answers, and those hidden reasoning tokens are billed as output tokens even though you never see them. Capping max_tokens and asking for short answers keeps them in check.
Is there a cheaper GPT model that is good enough?
Often. GPT-5.4 mini costs ₹74.05 per million input tokens and handles extraction, classification and routine drafting well. Try it on a sample of your real traffic before you decide.
Can I call GPT-5.5 without an OpenAI account?
Yes. With a Rapid Router key you call openai/gpt-5.5 through the OpenAI SDK by changing the base URL, and pay from one balance that also covers Claude, Gemini and 600+ other models.