Three different things on one page
The phrase apertus ai vs chatgpt compares objects of different kinds, so start by naming them.
ChatGPT is a chat product from OpenAI. You use it through an app or website, and the company decides which models run behind it, what the interface offers and which requests it declines. OpenAI also sells API access separately.
Apertus is an open language model from Swiss research institutions. Its weights are published, which means anyone can download them, run them on their own hardware or offer them through a hosting service. It is a model, not a single app, so how it behaves depends partly on who hosts it and how it is configured. For current details, read the project's own pages.
This service is a hosted, OpenAI-compatible text API for one open-weight model run by an independent provider and tuned to answer without content refusals for lawful adult use. It is not Apertus, not OpenAI, and not affiliated with either. We state that plainly because the site name may suggest otherwise.
Side-by-side comparison
| ChatGPT | Apertus | This API | |
|---|---|---|---|
| What it is | Chat product from OpenAI | Open model from Swiss institutions | Hosted API for one open-weight model |
| How you use it | App or website; separate OpenAI API | Download and self-host, or a third-party host | OpenAI-compatible HTTP API with a bearer key |
| Who sets the rules | OpenAI | The model's tuning plus whoever runs it | The provider; tuned for no content refusals on lawful adult use |
| Hard limit | Its usage policies | Depends on host | No sexual content involving minors |
| Pricing | Check OpenAI's current pricing | Free weights; you pay for hardware or a host | $0.15 / 1M input, $0.50 / 1M output, prepaid |
| Output types | Check current features | Text model | Text only |
I have left other products' prices out on purpose. They change, and a table that goes stale is worse than none. Look at each vendor's pricing page on the day you decide.
Where the three really differ
Control
With ChatGPT you get a polished product and few knobs. With an open model you can pick the hardware, the system prompt, the sampling settings and, if you host it yourself, every part of the stack. That freedom comes with work: serving a large model takes GPUs and operations time.
Handling of sensitive prompts
Every hosted assistant applies some policy. ChatGPT applies OpenAI's, and it will refuse some requests. An open model's behaviour depends on its tuning and on the host. This API is tuned to answer without content refusals for lawful adult use, which suits fiction, roleplay, companion apps and adult-oriented tools. The one fixed limit is that sexual content involving minors is never acceptable.
Where your data goes
With a consumer chat product, your conversation lives in that company's service under its terms. If you self-host an open model, it stays on your machines. On this API, prompts are not used for training. Read the privacy page before you send anything sensitive.
When each one is the better choice
- Pick ChatGPT when you want a general assistant app with a finished interface and you are happy with its content rules.
- Pick Apertus when openness and transparency matter most, when you want to inspect or run the model yourself, or when your organisation values a public-research origin.
- Pick this API when you are building your own app and want to call a text model over HTTP with no hardware to run, a published price, and no content refusals for lawful adult use.
Mixing is normal. Plenty of teams use a closed product for general work and a separate, more permissive endpoint for the part of their product that the first one declines.
Limits of this API, so there are no surprises: one model only, text only, no embeddings, no fine-tuning, a 100,000-token context window shared by prompt and completion. If you need any of those, look elsewhere or combine services.
Test all three on your own prompts
Benchmarks and blog posts tell you less than twenty prompts from your real workload. Write them down, run them everywhere and read the answers. Here is a small script for this API; swap the base URL and key to run the same loop against any other OpenAI-compatible endpoint.
from openai import OpenAI
client = OpenAI(
base_url="https://api.apertusaichat.com/v1",
api_key="YOUR_API_KEY",
)
prompts = [
"Write the opening scene of a noir story set in a rainy port city.",
"Explain how a SQL index speeds up a query, in four sentences.",
"Draft a firm but polite late-payment reminder for a client.",
"Traduis en francais : the meeting moved to Thursday morning.",
]
for p in prompts:
r = client.chat.completions.create(
model="uncensored",
messages=[{"role": "user", "content": p}],
temperature=0.7,
max_tokens=300,
)
print("PROMPT:", p)
print(r.choices[0].message.content)
print("-" * 40)
Score each answer for accuracy, tone, length and whether it refused. Include a few prompts in languages you care about; multilingual quality varies a lot between models and is easy to check.
Questions to settle before you choose
Four questions cut through most of the noise.
- Do I need a finished interface? If yes, you want a product, not an API. ChatGPT is one; an API is a building block.
- Must I run it where I control the data? If yes, an open model on your own hardware is the straightforward answer, and Apertus is one candidate.
- Will my users ask for things a general assistant declines? If yes, a model tuned for lawful adult use removes a recurring support problem. If no, you do not need that feature, and a conventional assistant is simpler.
- How many tokens will I move? Multiply requests by tokens and by the rate. At $0.65 for one million input plus one million output tokens, this API is easy to price, but check others on the day you decide.
Write the answers down before you compare features. Teams that skip this tend to pick the tool with the longest feature list and then work around it.
Cost and access for this API
Billing is prepaid credit with no subscription, and credit never expires. Failed or refused requests are not charged. A request with 2,500 input tokens and 400 output tokens costs 2,500 x $0.15 / 1M = $0.000375 plus 400 x $0.50 / 1M = $0.0002, which is $0.000575. One thousand of those cost $0.575.
Top-ups run from $20 to $500 in whole dollars, paid with USDT on TRC20 or USDC on Base, with a bonus of 5% from $50, 10% from $100 and 15% from $200. See the pricing page for the details.
New accounts can claim $0.50 of trial credit with one click on the API key page. It needs no card and is valid for 7 days. The trial key allows 2 concurrent requests and 60 per minute; a full key allows 8 and 300. The docs cover streaming, tool calling and JSON mode. For production concerns, the production checklist is a good next read.
