Myth: We are a proxy for GPT-4
Our API serves a distinct open-weight model, not a relay for OpenAI’s GPT-4. When you send requests to https://api.nsfwqwen.com/v1, you are interacting with our own servers running a specific large language model tuned for uncensored responses. It is not GPT, Claude, or Gemini. We do not route your traffic through another vendor. This means the behavior you see is consistent with our specific tuning, not the ever-changing policies of a third-party provider. You get direct access to our infrastructure without hidden middlemen.
Myth: All prompts are used for training
Your data remains yours. We do not use your prompts or completions to train our base model. This is a critical distinction for developers handling proprietary characters or sensitive roleplay scenarios. When you send a request, the tokens are processed for inference and then discarded. There is no hidden data mining. Your context window of 100,000 tokens is used solely for that specific conversation turn. You retain full privacy over your creative content.
Myth: Unlimited requests
There are hard limits to ensure stability for everyone. You get 300 requests per minute per key. Concurrent connections are capped at 8 per key. Request bodies cannot exceed 8 MB. These limits prevent a single user from monopolizing resources. If you need higher throughput, you can generate additional keys, though only one active key is allowed per account at a time. This structure balances cost efficiency with reliability. You can scale by managing keys rather than waiting for arbitrary throttling.
Myth: We support image or audio generation
We focus strictly on text. The endpoints are POST /v1/chat/completions and GET /v1/models. There are no embeddings, no image generation, no audio, and no video processing. We do not offer fine-tuning or OCR services. This specialization keeps our latency low and our pricing simple. If you need multimodal features, you must use a different provider. Our uncensored LLM excels at long-context dialogue, not pixel manipulation. Keep your stack modular.
Myth: You can get refunds on credit
Credit is non-refundable once purchased. The system charges you for real token usage. Errors and refusals are free, but your prepaid balance is not returned. This model works because credit never expires. You can top up whenever you want. If you make a mistake like a double charge, we fix it via Support. There are no monthly fees to lose. Your balance sits there until you use it. It is a straightforward prepaid utility.
Myth: It is a chat app for users
This is an API for developers. There is no built-in chat website or mobile app for end-users. You integrate our API into your own front-end or script. The interface is JSON. You handle the UI. This gives you complete control over the user experience. You can build a custom NSFW chatbot interface that fits your brand perfectly. The tool is the API, not a consumer app. Developers love this because it is flexible.
Myth: We offer an SLA or uptime guarantee
We do not publish SLA percentages or uptime guarantees. There are no SOC2 or HIPAA certifications. We are a lean service focused on performance and price. You get reliable access, but no contractual guarantee of 99.9% availability. This keeps costs low. For most roleplay and scripting use cases, this is perfectly acceptable. You trade corporate guarantees for lower prices and direct model access. It is a transparent trade-off.
Myth: It is free forever
The service is not free. You pay $0.15 per 1M input tokens and $0.50 per 1M output tokens. Errors are free. You get $0.50 in trial credit for new accounts, but that expires in 7 days. After that, you need to top up. You can pay with crypto only. Prices are transparent. There are no hidden subscription fees. You only pay for what you use. This model scales with your actual demand.
Myth: We are the original Qwen model
We are not Qwen. We are not OpenAI. We are an independent service running our own uncensored open-weight model. We do not serve the original Qwen model from Alibaba. We do not serve GPT-4 or Claude. Our model is tuned specifically to answer without content refusals for lawful adult use. It is a different architecture with different behavior. We use the name to describe our focus, not our lineage. The model ID is simply uncensored. Check our docs for specific capabilities.
