GPT-5 mini API: the small tier, priced live
The GPT-5 mini API and the rest of the small tier, gpt-5-nano, gpt-5.4-mini and gpt-5.4-nano, on one key, priced per million tokens and per 1,000 chat turns, read live next to OpenAI's list. Paid by card or in USDT, USDC, BTC, ETH, SOL, no KYC, nothing added on the top-up.
Mini and nano, per million tokens
USD per million tokens, read from the catalog at render. OpenAI's pricing page is the source of the list column where the catalog has one.
| Model | Input | Cached | Output | Developer list in / out | Context |
|---|---|---|---|---|---|
| GPT 5 Mini | $0.80 | $0.08 | $6.40 | — | 400K tokens |
| GPT 5 Nano | $0.2376 | $0.0238 | $1.9008 | — | 400K tokens |
| GPT-5.4 Mini | $0.75 | $0.075 | $4.50 | $0.75 / $4.50 | 400K tokens |
| GPT 5.4 Nano | $0.1335 | $0.0134 | $0.8346 | — | 400K tokens |
USD per million tokens, live. 'Developer list' is the catalog's record of OpenAI's own price. The GPT price hub has every GPT id on the key.
50,000 chat turns, costed
A chat turn is 2,000 tokens in (a system prompt, some history, the question) and 500 out. A classification call is 600 in and 20 out. Uncached, at today's price.
| Model | Per 1,000 chat turns | Per 1,000 classifications | 50,000 turns a month | Same month at list |
|---|---|---|---|---|
| GPT 5 Mini | $4.80 | $0.608 | $240.00 | — |
| GPT 5 Nano | $1.43 | $0.1806 | $71.28 | — |
| GPT-5.4 Mini | $3.75 | $0.54 | $187.50 | $187.50 |
| GPT 5.4 Nano | $0.6843 | $0.0968 | $34.22 | — |
A bot that repeats the same system prompt pays the cached rate on it where the model supports caching, which lowers the input side of every row.
A key in three steps, then the first request
- 01Open an accountAn email address. No OpenAI account, no identity documents, no phone number.
- 02Top upFrom $10 by card, Apple Pay, PayPal or in USDT, USDC, BTC, ETH, SOL. USD balance, no expiry, no subscription.
- 03Create a keyIn the console, with a monthly spend limit if the bot runs unattended. Use gpt-5-mini or gpt-5-nano as the model id.
Base URL https://api.relayswitch.io/v1, your key (it starts with sk-relay-), the model id. Nothing else changes.
from openai import OpenAI client = OpenAI(base_url="https://api.relayswitch.io/v1", api_key="sk-relay-...")reply = client.chat.completions.create( model="gpt-5-mini", messages=[{"role": "user", "content": "Say hello in five words."}],)print(reply.choices[0].message.content)curl https://api.relayswitch.io/v1/chat/completions \ -H "Authorization: Bearer $RELAYSWITCH_KEY" \ -H "Content-Type: application/json" \ -d '{"model": "gpt-5-mini", "messages": [{"role": "user", "content": "Hello"}]}'OpenAI's GPT-5 mini model page documents the limits. Swap the id for gpt-5-nano on the routing and classification calls and keep mini for the replies a person reads; both bill to the same balance.
What is stored: the email address, the top-ups, the keys, and one usage record per request (request id, model, token counts, charge). Not stored: request and response bodies, a name, an ID, a phone number, or a card for crypto top-ups. The privacy page has the list.
Mini, nano, or the full-size model
gpt-5-nano for the calls where the output is a label, a route, a yes or no, or a field pulled out of a record. Thousands of them a day cost what one long conversation costs on a flagship.
gpt-5-mini for the replies a person reads: support answers, summaries, rewrites, first-pass drafts. The month table above is the budget line for a bot built on it.
gpt-5.2 and up when the small tier gets the answer wrong often enough to cost more in escalations than the price gap. The GPT-5.2 API page has the same tables for the full-size tier; a fallback is one string change on the same key, so start on mini and promote the prompts that need it.
GPT-5 mini by API: questions
Price per turn, mini against nano, the key.
The small tier on a prepaid key
From $10 by card or in USDT, USDC, BTC, ETH, SOL. No OpenAI account, no KYC.