Key points
- Dolphin is a family of community fine-tunes, so there is no single official API to sign up for.
- You can serve one yourself on a local machine, serve it on a rented GPU, or call a hosted endpoint.
- A hosted OpenAI-compatible endpoint needs only a base URL, a key and the model ID uncensored.
- Hosted pricing here is $0.15 per 1M input tokens and $0.50 per 1M output tokens, with no subscription.
What people mean by a Dolphin model
Dolphin is the name of a line of community fine-tunes of open-weight base models. The tuning data is filtered so the model follows instructions without lecturing or refusing ordinary requests. Because the base model changes between releases, there is no single Dolphin vendor with one official endpoint. When someone asks for a dolphin llm api, they usually mean any API that serves an uncensored model of that style.
That matters for how you plan. You are not choosing a brand, you are choosing where the weights run and who operates them. The model quality depends on the base model and the tune, and the operational side depends on the route you pick below.
One note on scope: these models are tuned for lawful adult use. No provider here, and no sensible one anywhere, will serve sexual content involving minors. Treat that as a fixed boundary in your own application as well.
Route 1: run it on your own machine
The cheapest route per token is a model you already own hardware for. Tools such as Ollama and llama.cpp download quantized weights and expose a local server. Both can offer an OpenAI-style endpoint on localhost, so your client code looks the same as it would for any hosted service.
The catch is memory. A quantized model needs roughly its file size in RAM or VRAM, plus room for the context cache. Small models fit on a laptop, but larger ones need a GPU with a lot of memory, and long prompts make the cache grow. If the model does not fit in VRAM, it spills to system RAM and speed drops sharply.
Local is a good fit when you work offline, when data must never leave your machine, or when you want to experiment with different fine-tunes. It is a poor fit for a product with users, because your laptop becomes the server. For a full walkthrough of the local side, see the separate article on how to run a Dolphin-style model.
Route 2: rent a GPU and serve it yourself
Renting a GPU by the hour gives you bigger models without buying hardware. You install a serving stack such as vLLM or llama.cpp's server, load the weights, and expose a port. You then own everything around it.
- Setting up drivers, the serving software and TLS.
- Adding authentication so strangers cannot use your GPU.
- Restarting after crashes and watching memory.
- Paying for the hours the box is on, whether or not anyone sends requests.
That last point is the real cost driver. A GPU that idles overnight still bills. The route pays off when you have steady, high traffic and someone willing to operate it. For a side project with bursts of use, you pay mostly for idle time. Check the current hourly rates on the provider's own pricing page, since they change.
Route 3: call a hosted OpenAI-compatible endpoint
A hosted endpoint removes the operating work. You send HTTPS requests and pay for the tokens you use. This site offers that: a hosted, OpenAI-compatible text API for a single open-weight model, run by an independent provider and tuned to answer without content refusals for lawful adult use. To be clear about what it is not: it is not the official API of any model vendor, and it is not a router across several models. The model ID is uncensored.
Because the API is OpenAI-compatible, any OpenAI SDK works once you change two settings, the base URL and the key. Here is a minimal Python call:
from openai import OpenAI
client = OpenAI(
base_url="https://api.dolphinllm.com/v1",
api_key="YOUR_KEY",
)
resp = client.chat.completions.create(
model="uncensored",
messages=[
{"role": "system", "content": "You are a direct, helpful assistant."},
{"role": "user", "content": "Explain how a context window works."},
],
)
print(resp.choices[0].message.content)The same request in curl:
curl https://api.dolphinllm.com/v1/chat/completions -H "Authorization: Bearer YOUR_KEY" -H "Content-Type: application/json" -d '{"model":"uncensored","messages":[{"role":"user","content":"Hello"}]}'Streaming (stream: true), tool calling, JSON mode and the usual sampling parameters work as in the OpenAI API. The context window is 100,000 tokens, prompt plus completion. The API documentation lists every supported endpoint and parameter.
Side-by-side comparison
| Factor | Local | Rented GPU | Hosted endpoint |
|---|---|---|---|
| Upfront work | Install and download | Server setup, auth, TLS | Get a key |
| Cost model | Hardware you own | Hourly, idle time included | Per token, prepaid |
| Model size | Limited by your VRAM | Limited by the GPU you rent | One model, fixed |
| Model choice | Any fine-tune you can download | Any fine-tune you can load | Only the model offered |
| Data location | Stays on your machine | On the rented box | Sent to the provider |
| Maintenance | You | You | Provider |
The row that decides most cases is model choice. If you need one specific Dolphin tune, only the first two routes give you that. If you need a capable uncensored model behind a stable API and you do not care which exact tune it is, the hosted route saves you the operations work.
What a hosted call costs
Pricing is $0.15 per 1M input tokens and $0.50 per 1M output tokens, paid from prepaid credit. There is no subscription, failed or refused requests are not charged, and credit never expires. A typical chat turn with 2,000 input tokens and 500 output tokens costs:
- Input: 2,000 x $0.15 / 1,000,000 = $0.0003
- Output: 500 x $0.50 / 1,000,000 = $0.00025
- Total: $0.00055 per request, or $0.55 per 1,000 requests
Top-ups are any whole amount from $20 to $500, with +5% from $50, +10% from $100 and +15% from $200. You can pay with crypto (USDT on TRC20, USDC on Base). The pricing page has the current details.
To compare against a GPU you rent, divide the hourly rate by the number of requests you can serve in an hour at your real traffic. At low or bursty volume the per-token model usually wins, because you never pay for idle time.
Trying it before you pay
New users can claim $0.50 of trial credit with one click on the key page after sign-up. It is one per person, valid for 7 days, and needs no card. Sign-up is with Google or with email and password. At the prices above, $0.50 covers about 900 of the 2,000-in, 500-out requests from the example, which is plenty to test whether the model's tone fits your use. Trial keys are limited to 2 concurrent requests and 60 requests per minute.
Get your key on the API key page. If you want a model with no login at all, see the no-login LLM article, and for client setup in a chat frontend read about Open WebUI endpoints.
