An uncensored LLM API

An uncensored LLM API is a hosted HTTP endpoint for a model that completes prompts a stock chatbot refuses. Refuseless is one: base URL https://api.refuseless.com/v1, models glm-5.3-abliterated and glm-5.3-flash-abliterated, OpenAI and Anthropic request shapes, and prompts that are never saved or used for training. Auth is a bearer API key. POST /v1/chat/completions is the usual call, and GET /v1/models lists the ids your key can use.

The same search also turns up pages promising "unfiltered AI," "no-filter AI," "no-restriction AI," and "unrestricted AI chat." Those are the same query wearing different adjectives. This is the page for it. The model behind the API is abliterated, which means the refusal change is in the weights. If you need the definition of that word before the endpoint, read what an uncensored LLM is. To compare this API with abliteration.ai, Venice, and OpenRouter, read best uncensored LLM APIs in 2026.

How do you call an uncensored LLM API?

A call is an OpenAI chat-completions body. Set model to a lineup id: glm-5.3-abliterated for GLM 5.3 Abliterated, or glm-5.3-flash-abliterated for GLM 5.3 Flash Abliterated. The models page lists both. A wrong id comes back as HTTP 404 with code model_not_found. A missing key comes back as HTTP 401, "Missing bearer API key." A wrong key comes back as HTTP 401, "Invalid API key."

curl https://api.refuseless.com/v1/chat/completions \
  -H "Authorization: Bearer $REFUSELESS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3-abliterated",
    "messages": [{"role": "user", "content": "Name three logs you would read first on a suspected token theft."}]
  }'

What other endpoints does the API serve?

The same host also accepts POST /v1/messages in the Anthropic Messages shape, so the Anthropic SDK works with the same key. It also serves POST /v1/responses, POST /v1/completions, and POST /v1/embeddings where the model offers them. An unknown path returns HTTP 404 with the list of supported paths in the message. There is no separate "uncensored mode" flag. You do not pass a query parameter to turn refusal off. The checkpoint is already the abliterated one.

Does the API keep prompts, and what does it cost?

The published retention line: prompts and outputs are never saved after the request is served, and they are not used to train models. The privacy policy also says that some requests may be served by third-party model providers, which receive the content without your IP address, email, account id, or API key. That is not a data-processing addendum, and this page will not invent a region, a subprocessors list, or a deletion timer the product has not published. Read zero prompt retention for how that phrase differs from "we don't train" and "ephemeral processing."

Price is set per model and billed per token. Per million tokens on the pricing page: GLM 5.3 Abliterated, $1.00 in, $0.15 cached, $2.00 out; GLM 5.3 Flash Abliterated, $0.12 in, $0.025 cached, $0.50 out; DeepSeek V4.1 Flash 1017 Abliterated, $0.12 in, $0.025 cached, $0.45 out. Settlement methods published there are Visa, Mastercard, BTC, ETH, USDT, USDC, and XMR. The Python and Node pages show the base-URL change and nothing else, because nothing else in a normal OpenAI SDK call has to move.