AI API Key Free Trial: What $0.50 Credit Buys

An ai api key free trial here means $0.50 of credit, claimed with one click after sign-up, with no card and a 7-day window. At current prices that is roughly 1,500 short chat turns or about 150 document summaries.
Key points
- The trial is $0.50, one per person, valid for 7 days, no card needed.
- At $0.15 per 1M input and $0.50 per 1M output, one chat turn costs a fraction of a cent.
- $0.50 covers about 1,560 short chats or about 150 summaries of 20,000-token documents.
- Trial keys allow 2 concurrent requests and 60 per minute, so plan batch jobs accordingly.
How the free trial works
Sign up with Google, or with an email and password. Open the API key page and press the trial button. $0.50 of credit lands on your account immediately. You do not enter a card, and the credit is limited to one claim per person.
The credit is valid for 7 days. After that it lapses, so decide early what you want to test. Credit you buy later behaves differently: prepaid top-ups never expire.
A trial key is a normal key with tighter limits: 2 concurrent requests and 60 requests per minute, against 8 and 300 for a full key. Everything else is identical, including the model ID uncensored, streaming, tool calling and JSON mode.
The price list the math uses
Billing is per token, prepaid, with no subscription. Input and output are priced separately:
| Token type | Price per 1M | Price per 1,000 |
|---|---|---|
| Input (your prompt) | $0.15 | $0.00015 |
| Output (the reply) | $0.50 | $0.0005 |
Failed or refused requests are not charged. That matters in a trial, because your first attempts often contain a typo in the model name or a malformed body, and those cost nothing.
Two edge cases frame the budget. If you only sent input tokens, $0.50 would buy $0.50 / $0.15 = 3.33M of them. If you only generated output, it would buy 1M tokens. Real work sits between the two, which is why the examples below count both sides.
What $0.50 buys: worked examples
Short chat turns
Take a typical exchange of 800 input tokens (a system prompt plus a question) and 400 output tokens.
- Input: 800 / 1,000,000 x $0.15 = $0.00012
- Output: 400 / 1,000,000 x $0.50 = $0.0002
- Per turn: $0.00032
- $0.50 / $0.00032 = 1,562 turns
Summarising documents
A 20,000-token document in, a 500-token summary out.
- Input: 20,000 / 1,000,000 x $0.15 = $0.003
- Output: 500 / 1,000,000 x $0.50 = $0.00025
- Per document: $0.00325
- $0.50 / $0.00325 = 153 documents
Filling the whole context window
The context window is 100,000 tokens, prompt plus completion. A 99,000-token prompt with a 1,000-token answer costs $0.01485 + $0.0005 = $0.01535. That is about 32 such calls from $0.50. Useful for testing long-document behaviour, expensive as a habit.
A small tool-calling agent
Say a run takes five model calls, each about 3,000 input tokens and 300 output tokens. Input is 15,000 tokens ($0.00225), output is 1,500 tokens ($0.00075), total $0.003 per run. The trial covers about 166 runs.
A quick way to burn the trial on purpose
The best use of a trial is a real test of your own workload, not a hello-world. Pick ten representative prompts, run them, and read the usage object each response returns. That gives you your true average tokens per request instead of a guess.
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["AIAPI_KEY"],
base_url="https://api.aiapisource.com/v1",
)
prompts = ["Draft a product description for a camping stove.",
"Explain what a webhook is in two paragraphs."]
total_in = total_out = 0
for p in prompts:
r = client.chat.completions.create(
model="uncensored",
messages=[{"role": "user", "content": p}],
)
total_in += r.usage.prompt_tokens
total_out += r.usage.completion_tokens
cost = total_in / 1e6 * 0.15 + total_out / 1e6 * 0.50
print(total_in, total_out, round(cost, 6))
Multiply the printed cost by your expected monthly request count and you have a budget. The pricing page lists the same rates if you want to cross-check.
Staying inside the trial limits
Because the trial key allows only 2 concurrent requests, a script that fires ten requests in parallel will get HTTP 429 with the code concurrency. Cap your worker pool at 2, or run requests one after another. Over 60 requests in a minute returns rate_limited instead. Both are safe to retry after a short pause, and rejected requests are not billed.
Streaming deserves a mention here. A streamed request occupies its concurrency slot until the last token arrives, so two long streams can block everything else on a trial key. If you see unexplained 429 responses, count your open streams first.
The balance is always visible. Each response includes an X-Balance-USD header, and GET /v1/balance returns it on demand, so you can watch the $0.50 go down in real time. The docs cover both endpoints.
Mistakes that waste trial credit
Most of the credit people lose in a trial goes to a handful of habits, not to the price.
- Resending long history. Chat APIs are stateless, so every turn resends the whole conversation as input. A 30-turn chat with 6,000 tokens of history costs 6,000 / 1,000,000 x $0.15 = $0.0009 of input on the last turn alone. Trim old turns or summarise them.
- Leaving max_tokens unset in loops. Output is the expensive side at $0.50 per 1M, and a model that rambles in a test loop spends more than three times as much per token than one that reads. Set a cap while experimenting.
- Testing with huge pastes. One 100,000-token prompt costs as much as roughly 47 short turns ($0.015 against $0.00032).
- Retrying blindly. A 429 is not billed, but a retry storm against the limit wastes your own time. Add a short sleep.
Keep a running tally from the usage object and you will know within a few requests whether your workload fits the trial or needs a top-up.
What to do after the trial
If the test went well, top up with any whole amount from $20 to $500. Larger top-ups add a bonus: +5% from $50, +10% from $100 and +15% from $200. A $50 top-up therefore gives $52.50 of credit, and $200 gives $230. You can pay by crypto (USDT on TRC20, USDC on Base).
For scale: $20 at the 800-in, 400-out chat profile above covers 20 / 0.00032 = 62,500 turns. There is no monthly fee underneath that, and the credit never expires, so an unused balance simply waits.
If the trial showed the model is not what you need, you lose nothing. Prompts are not used for training, and there is no subscription to cancel.
More guides
Questions and answers
Do I need a credit card for the free trial?
No. The $0.50 trial credit needs no card. Sign up with Google or email and password, then claim it with one click on the API key page.
How long does the trial credit last?
It is valid for 7 days from the claim. Credit you buy afterwards does not expire.
Can I claim the trial more than once?
No, it is one per person. If you need more room for testing, a $20 top-up is the smallest paid step.
Is the free trial key different from a paid key?
It works the same way, with lower limits: 2 concurrent requests and 60 per minute instead of 8 and 300.
Your key is one form away
Create an account, copy the key, change the base URL. That is the whole setup.