Why My Request Is Refused (It's Not the Model)
Requests refuse only when the content involves sexual acts with minors. This is the single hard limit that always applies to our uncensored LLM. Everything else—controversial politics, explicit fiction, or raw adult themes—passes through without filters. The model does not refuse based on tone, style, or general NSFW topics. It refuses only that specific category. This ensures consistent abliterated behavior for lawful adult use. You can send detailed, uncensored prompts without worrying about hidden policy gates. The refusal is binary and predictable. Check your payload for minor-related sexual content. If that is absent, the request will succeed. This makes the API reliable for NSFW uncensored applications. Do not assume general adult content triggers a refusal. Only the minor limit is enforced. This clarity removes guesswork from your integration. You get raw text output without arbitrary moderation. The model is tuned for this specific behavior. It does not mimic the safety layers of other vendors. This is a dedicated uncensored AI service. The refusal reason is straightforward. It is not a model error. It is a hard rule. This simplifies error handling in your code. You only need to check for that one refusal type. Everything else is fair game. The model answers directly. There are no soft filters. This is the core value of our API. It is predictable. It is transparent. It is uncensored.
Understanding the 100k Context Window Limit
The context window spans 100,000 tokens for both prompt and completion combined. This limit includes all previous messages in the conversation history. You must manage memory carefully to avoid exceeding this cap. The model does not automatically truncate older messages. Your application must handle history management. Send only the necessary context to stay within the limit. This ensures efficient token usage and lower costs. The 100k window allows for long-form analysis or extended dialogues. It supports complex reasoning tasks that require deep context. Monitor your token counts closely. Exceeding the limit causes an error. You must truncate or summarize older messages. This is a standard constraint for large context models. Our API enforces this strictly. There is no automatic fallback. You control the history. This gives you flexibility. You can keep long conversations. You can also start fresh. The choice is yours. The limit is clear. It is 100,000 tokens. Plan your architecture accordingly. This is a key feature for deep context needs. It supports extensive document processing. It allows for detailed role-playing. The limit is firm. Respect it for optimal performance.
Max Tokens: The 16k vs 2k Default Trap
Set max_tokens to 16,000 to get the full output capacity. The default limit is 2,048 tokens. This default often truncates long responses unexpectedly. You must explicitly define the max tokens in your request. Without this setting, your uncensored AI cuts off mid-sentence. This is a common pitfall for new integrations. The model can output up to 16,000 tokens per request. This allows for detailed essays, code blocks, or extended dialogues. Adjust this parameter based on your use case. Short queries need fewer tokens. Long analyses need more. Always specify the limit to avoid surprises. The default is conservative. The maximum is generous. Use the maximum for comprehensive answers. This ensures you get the full response. It is a simple configuration change. It prevents data loss. It improves user experience. The uncensored LLM is capable of long outputs. Trust its ability. Just give it the room to breathe. Set the limit high. Let the model perform. This is a critical setting. Do not overlook it.
Streaming: How to Capture Token Usage Data
Enable streaming via Server-Sent Events (SSE) to get real-time responses. Token usage data appears only in the final SSE chunk. You must parse the last event to get the total tokens. This is standard for OpenAI-compatible APIs. The uncensored AI online API follows this pattern. Stream the output for better UX. Parse the final chunk for billing metrics. This ensures accurate cost tracking. The token count includes both input and output. It reflects the actual resources used. Streaming reduces perceived latency. Users see text as it generates. The final chunk confirms the request is complete. It also contains the cost data. You need to handle this data in your client. Store it for your records. This is essential for prepaid credit management. Our API provides this data accurately. It helps you monitor usage. It ensures you pay only for what you use. Errors and refusals are free. This streaming pattern is efficient. It is reliable. It is standard. Use it for the best experience.
Function Calling Errors in Uncensored Models
Function calling works but requires strict JSON formatting. The uncensored model may hallucinate tool names or arguments. Validate the tool response before execution. Use response_format with json_object for stability. This forces the model to output valid JSON. It reduces parsing errors. Function calling is a powerful feature. It allows the model to interact with external systems. Use it for data extraction or action triggers. Ensure your tool definitions are clear. The model follows instructions better with precise definitions. Test your functions thoroughly. Errors usually stem from malformed JSON. Check the model output. Adjust your prompt if needed. The uncensored AI supports tools. It is compatible with the OpenAI standard. Use it for dynamic interactions. This adds versatility to your application. It is a key capability. Do not skip validation. It saves debugging time. It ensures reliability. The model is capable. Just guide it correctly.
JSON Mode Formatting Issues
Enable JSON mode by setting response_format to {"type": "json_object"}. This forces the model to output valid JSON strings. It prevents markdown code blocks around the JSON. The uncensored LLM adheres to this format. It ensures your parser receives clean data. Without this setting, the model may add explanatory text. This breaks JSON parsing. Always use JSON mode for structured data. It improves reliability. It reduces post-processing needs. The uncensored AI online API supports this feature. It is essential for function calling. It is also useful for data extraction. Use it for consistent outputs. It is a simple toggle. It makes a big difference. Ensure your client handles the JSON response correctly. Parse it safely. Handle errors gracefully. This feature is standard. It is well-supported. Use it for production apps. It ensures data integrity. It is a best practice. Do not skip it.
Crypto Top-Up Delays and Bonus Calculation
Top-ups are processed via crypto only. We accept USDT on TRC20 or USDC on Base. Payments from $20 to $500 are supported. You receive bonus credit based on the amount. A $50 top-up gives a 5% bonus. A $100 top-up gives a 10% bonus. A $200 top-up gives a 15% bonus. The bonus is applied immediately. Credit never expires. There are no monthly fees. Errors and refusals are free. You pay only for real token usage. The prepaid model is transparent. No hidden costs. No subscription lock-in. Credit is topped up directly. Use the Support page for double charges. The process is straightforward. It is efficient. It is secure. Use crypto for fast settlement. It avoids bank delays. It is the preferred method. It is the only method. Stick to it.
Rate Limits: 300 RPM and Concurrency Rules
The API allows 300 requests per minute per key. You can have 8 concurrent requests at the same time. The request body is limited to 8 MB. These limits ensure fair usage. They prevent abuse. Exceeding the rate limit returns an error. Handle retries with backoff. The uncensored AI online API enforces these limits strictly. They are standard for high-performance APIs. Monitor your usage. Stay within the limits. If you need higher limits, contact support. The limits are per key. One key per account. Replace the key if needed. This ensures security. It is a simple model. It is effective. It protects the service. Use it wisely. It ensures stability. It guarantees performance. Respect the limits. They are there for a reason. They keep the API fast. They keep it reliable. Follow the rules.
When to Stop Using Uncensored AI
Stop using this API if you need image, audio, or video generation. We only provide text output. We do not offer embeddings or fine-tuning. If you need model routing or multiple vendors, this is not for you. We run a single open-weight model. It is tuned for uncensored output. It is not a general-purpose aggregator. If you need SLA guarantees or SOC2 compliance, look elsewhere. This is a developer-focused API. It is for raw text generation. It is for NSFW uncensored use cases. It is for developers who want control. It is not for end-users who want a chat website. It is an API. Use it for integrations. Use it for apps. Use it for data. It is powerful. It is flexible. It is uncensored. But it is not everything. Know its limits. Use it where it fits. It is a tool. Use it well.
