Dolphin AI: Production Guide
Dolphin AI models offer a distinct approach to uncensored language generation, providing developers with open-weight options that bypass the content filters of major providers. This guide explains the technical differences between abliterated and fine-tuned models, details integration with standard OpenAI-compatible SDKs, and outlines cost-effective prepaid billing for high-volume deployments.
Last updated
Summary
- Dolphin models are fine-tuned on new datasets, while abliterated models have their refusal mechanisms deleted, offering different styles of uncensored behavior.
- Our API serves a single uncensored model via the standard /v1/chat/completions endpoint, ensuring compatibility with existing client code.
- Prepaid token billing starts at $0.15 per million input tokens, with crypto top-ups offering bonus credits for larger amounts.
- The service supports a 100,000-token context window and streaming responses, making it suitable for long-context applications.
Why Choose an Uncensored LLM for Production
Uncensored models remove the polite refusals that standard LLMs apply to creative, adult, or controversial topics. Developers choose this path when they need raw text generation without pre-filtering logic interfering with their application's specific use case.
Major providers often filter content at the API level, which can break workflows that rely on explicit language or niche topics. An uncensored LLM allows your application to handle content moderation locally, giving you full control over what appears to end users.
- Full creative freedom for fiction and roleplay.
- No unexpected refusals for lawful adult content.
- Consistent behavior across controversial topics.
Understanding Dolphin vs. Abliterated Models
The core difference lies in how the model achieves its uncensored state. Dolphin models are fine-tuned on new datasets, resulting in an "edgy" but coherent style. Abliterated models have the "refusal" part of the neural network deleted, preserving the original model's base knowledge.
Neither approach is universally superior. Fine-tuning can improve instruction following, while abliteration preserves the original model's voice. Both methods effectively remove content filters for lawful adult use.
| Method | Process | Result |
|---|---|---|
| Dolphin | Fine-tuning | Edgy, high-coherence text |
| Abliterated | Refusal deletion | Original model voice, no filters |
Integrating the OpenAI-Compatible Endpoint
Integration requires changing the base URL and API key in your existing OpenAI SDK configuration. Our API serves a single model with the ID "uncensored" via the standard endpoints.
This approach means you can use the same client code that works with GPT-4 or Claude. You simply point the SDK to our server. The API supports streaming, function calling, and JSON mode.
Base URL: HTTPS://api.dolphinaimodels.com/v1
Model ID: "uncensored"
Endpoints: POST /v1/chat/completions and GET /v1/models.
Managing Context Windows and Token Limits
The API supports a 100,000-token context window, which includes both prompt and completion tokens. This allows for long documents or extended conversations without immediate truncation.
However, the maximum output per request is 16,000 tokens. If you do not specify max_tokens, the default is 2,048 tokens. This limit ensures consistent performance for streaming responses.
Monitor your token usage closely. Errors and refusals do not consume credit, so you can test your prompts without cost.
Optimizing Costs with Prepaid Credit
Credit is charged by real token usage, with input tokens at $0.15 per million and output tokens at $0.50 per million. There are no monthly fees or subscriptions.
Crypto top-ups start at $20 and go up to $500. Bonus credits apply to larger amounts: +5% for $50, +10% for $100, and +15% for $200. Credit never expires.
New accounts receive $0.50 in trial credit, valid for 7 days. You can top up using USDT (TRC20) or USDC (Base).
Handling Streaming and Function Calling
Streaming is supported via Server-Sent Events (SSE). The last chunk of the stream includes token usage information, allowing you to track costs in real-time.
Function calling (tools) is fully supported. You can define tools in your request and let the model decide when to use them. JSON mode is also available for structured outputs.
Parameters like temperature, top_p, stop, and seed are all configurable. This flexibility allows you to tune the model's creativity and determinism.
Privacy and Data Usage Policies
Your prompts are not used for training. This is a critical distinction from many free or low-cost providers who use your data to improve their models.
An account requires only an email address. No phone number is needed. You can sign up with Google or use a standard email and password.
A hard content limit always applies: sexual content involving minors is refused. All other lawful adult content is allowed.
Troubleshooting Common API Errors
The most common error is hitting the rate limit. The API allows 300 requests per minute and 8 concurrent requests per key. If you exceed this, you will receive a 429 error.
Another issue is request body size. The maximum request body is 8 MB. If you are sending large contexts, ensure you are within this limit.
Errors are free. If a request fails due to a server issue or invalid input, you are not charged for those tokens.
Final Checklist for Deployment
Before going live, verify your API key is active. Ensure your base URL is set correctly to HTTPS://api.dolphinaimodels.com/v1.
Test your streaming implementation. Confirm that function calling works as expected with your chosen SDK.
Check your token usage limits. Ensure you have enough prepaid credit for your projected volume.
Remember that you can replace your API key at any time. Each account has one active key, which is replaced when you generate a new one.
Common questions
What is the difference between dolphin models and abliterated models?
Dolphin models are fine-tuned on new datasets, resulting in an edgy but coherent style. Abliterated models have their refusal mechanisms deleted, preserving the original model's base knowledge. Both methods effectively remove content filters for lawful adult use.
Does the API support streaming and function calling?
Yes, the API supports streaming via Server-Sent Events (SSE) and includes token usage in the last chunk. Function calling (tools) is fully supported, along with JSON mode for structured outputs.
How do I top up my credit?
You can top up using USDT (TRC20) or USDC (Base). Minimum amount is $20, and maximum is $500. Bonus credits apply: +5% for $50, +10% for $100, and +15% for $200.
Which payment methods do you accept?
Crypto only: USDT on TRC20 or USDC on Base, any whole amount from $20 to $500. Bonus credit applies from $50.
Unlock your API key
Sign up, retrieve your key, update your base URL. Complete in seconds.
Open account