What the trial is and who it is for
First, what this is not. The service on this site is an independent, OpenAI-compatible API that runs one open-weight model tuned to answer without content refusals for lawful adult use. It is not the official DeepSeek API, so its trial terms have nothing to do with whatever DeepSeek itself offers. If you want to know about the vendor's own free credits, check their current pricing and sign-up pages.
The trial here is a small amount of real credit, enough to find out whether the model and the API suit your project. It is meant for a quick test: wire it to your client, run your own prompts, check the answers and the speed, and decide.
The terms in full:
- $0.50 of credit, one claim per person.
- Claimed with one click on the API key page after you sign up.
- Valid for 7 days from the claim.
- No card needed.
- Sign-up is with Google, or with email and password.
How to claim it, step by step
- Open the API key page and sign up with Google, or create an account with an email address and password.
- Stay on the key page. After sign-up it shows a button to claim the trial credit. Click it once.
- Create or copy your API key from the same page.
- Make a first call, using the base URL
https://api.deepseekapipricing.com/v1and the modeluncensored.
A first call with curl, which also lists the model:
curl https://api.deepseekapipricing.com/v1/models \
-H "Authorization: Bearer $API_KEY"
curl -i https://api.deepseekapipricing.com/v1/chat/completions \
-H "Authorization: Bearer $API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "uncensored",
"messages": [{"role": "user", "content": "Give me three names for a lighthouse cat."}],
"max_tokens": 60
}'
The -i flag prints the response headers. Look for X-Balance-USD, which shows what credit remains after the call, and X-Request-Id, which is worth logging so you can trace a specific request.
What $0.50 buys in tokens
Pricing is $0.15 per million input tokens and $0.50 per million output tokens. Here is what the trial credit covers for different kinds of use. Each line is plain arithmetic from those two rates.
| Use | Tokens per request | Cost per request | Requests for $0.50 |
|---|---|---|---|
| Short chat turn | 1,000 in, 500 out | $0.00015 + $0.00025 = $0.0004 | 1,250 |
| Typical assistant reply | 1,500 in, 400 out | $0.000225 + $0.0002 = $0.000425 | about 1,176 |
| Document question | 20,000 in, 1,000 out | $0.003 + $0.0005 = $0.0035 | about 142 |
| Full 100,000-token prompt, 1,000 out | 100,000 in, 1,000 out | $0.015 + $0.0005 = $0.0155 | 32 |
As raw volume, $0.50 buys about 3.33 million input tokens (0.50 / 0.15) if you only read, or exactly 1 million output tokens if you only write. Most real use mixes both, so think in hundreds to a thousand requests, not millions of tokens. Failed or refused requests are not charged, so a call that errors out does not eat your credit.
That is enough for a serious evaluation. You can run a 200-prompt test set, a few long-context checks, and a streaming test, and still have credit left over.
Limits on a trial key
Trial keys are capped lower than full keys.
| Trial key | Full key | |
|---|---|---|
| Concurrent requests | 2 | 8 |
| Requests per minute | 60 | 300 |
Both share the same 100,000-token context window and the same 8 MB body limit. The model is the same too, and so are the features: streaming, tool calling and JSON mode.
In practice the caps mean a trial is good for interactive testing and short scripts, not for a bulk job. If a loop of calls goes over 60 per minute, you get HTTP 429 with the error code rate_limited. If you start three requests at the same moment on a trial key, the third returns 429 with the code concurrency. Remember that a streamed request holds its slot until the stream ends. The simple fix for both is a small client-side limiter:
import time
from openai import OpenAI
client = OpenAI(
base_url="https://api.deepseekapipricing.com/v1",
api_key="YOUR_TRIAL_KEY",
)
prompts = ["Name a color.", "Name an animal.", "Name a city."]
for p in prompts:
raw = client.chat.completions.with_raw_response.create(
model="uncensored",
messages=[{"role": "user", "content": p}],
max_tokens=20,
)
reply = raw.parse()
print(reply.choices[0].message.content, "| balance:", raw.headers.get("X-Balance-USD"))
time.sleep(1.1) # stays under 60 requests per minute
Sending one request at a time, one second apart, respects both limits.
What to test during the 7 days
Seven days passes quickly, so plan the test before you claim. A good list:
- Your refusal cases. Take ten prompts that your current model declines and that are lawful. See how this one handles them.
- Your quality bar. Run five tasks you know well, such as a summary, a rewrite or a code fix, and judge the output against what you would accept.
- Your client. Confirm that your SDK, framework or app connects with only a base URL and key change. Anything that speaks the OpenAI protocol should.
- Streaming and tools. If your product uses them, test them with your real schemas.
- Your token counts. Read the usage block on real calls and multiply out a monthly estimate.
Write the results down as you go: prompt, answer, tokens used and the balance header. A plain spreadsheet is enough. On day seven you will have a record that tells you whether to top up or walk away, and it costs nothing beyond the trial credit itself. If a result surprises you, rerun the prompt with a fixed seed and a lower temperature to see whether the behavior is stable before you judge it.
The documentation has copy-paste examples for each of these, and the DeepSeek V4 API page covers the same service from the key-and-pricing side.
Common first-day problems
Most trial problems come from a handful of causes. If your first call fails, check these in order.
- The credit is not there yet. The trial is claimed with a separate click on the key page after sign-up. Signing up alone does not add it.
- Wrong base URL. It must be
https://api.deepseekapipricing.com/v1, with the/v1on the end. Clients that append/chat/completionsthemselves need exactly that base. - Wrong model name. Use
uncensored. A vendor-style model ID will not match. - A leftover OpenAI key. Many SDKs read
OPENAI_API_KEYfrom the environment. Pass the key you got here explicitly so the right one is sent. - 429 responses. On a trial key you are limited to 2 concurrent requests and 60 per minute. Space your calls out, or reduce your worker count to 2 or fewer.
- A very long prompt. Prompt plus completion together must fit in 100,000 tokens, and the request body cannot exceed 8 MB.
If a request fails outright, it is not charged, so you can retry without watching your credit drop.
Checking your usage as you go
The cheapest way to understand your spend is to read it from the data. Every response carries a usage block, and the headers include X-Balance-USD. Log both for the first few hundred calls. After a day you will know your average input and output sizes, and you can multiply them by the two rates to project a month. A test that uses $0.10 of the $0.50 in an afternoon tells you something real: a thousand times that is the scale of a mid-sized project, and you can price it before you spend a dollar.
Why there is no unlimited free tier
Every token sent to a language model uses GPU time that someone pays for. An unlimited free tier would invite automated sign-ups and force tight limits and queueing on everyone. A small fixed credit, one per person, gives you a real test without that trade-off.
After the trial, the service works on pay as you go from prepaid credit, with no subscription. Credit never expires. You can top up any whole amount from $20 to $500. Bonuses apply on larger top-ups: 5% from $50, 10% from $100 and 15% from $200. You can pay with with crypto (USDT on TRC20 or USDC on Base).
For scale, $20 buys 133.3 million input tokens (20 / 0.15) or 40 million output tokens (20 / 0.50). A $100 top-up gives $110 of credit. If the trial shows the model works for you, the pricing page has the details.
