UncensoredLLM is a developer-focused gateway to uncensored, abliterated and roleplay-oriented open-weight models. One prepaid API key provides access through OpenAI-compatible endpoints, with additional support for the Anthropic Messages format. It is operated by Uplink Switch LLC and is intended for developers who want fewer model refusals while retaining familiar SDKs and coding-agent integrations.

What UncensoredLLM offers

The live catalogue listed more than 370 language models at the time of review, alongside a smaller selection of image and video models. The catalogue changes frequently and includes original open-weight builds, third-party fine-tunes and models described by the service as house builds. UncensoredLLM is the gateway and hosting layer, not necessarily the developer of the underlying weights.

  • OpenAI Chat Completions and Responses endpoints plus Anthropic-compatible messages
  • Model selection by ID, automatic routing and ordered fallbacks
  • Web search, web fetch, remote MCP servers and a code sandbox as agent tools
  • PDF and document input with optional redaction before a request leaves the gateway
  • Response caching, long-conversation handling and reusable presets
  • Session traces, CSV usage export and integrations with Langfuse, webhooks and OTLP
  • Embeddings, speech-to-text, reranking and batch endpoints on the same account
  • Signed request receipts and monthly statements for checking billed usage

Pricing

Prepaid access starts with a US$10 top-up and the paid balance does not expire. Each model has separate input and output rates per million tokens, while media and tool operations use their own published units. On 21 November 2026, the lowest listed text rate was Gemma 3n E2B at US$0.12 per million input tokens and US$0.08 per million output tokens. Rates extend much higher for larger or specialised builds, so the live model row matters more than the headline minimum.

Optional recurring usage plans sell a larger monthly usage allowance than their cash price: Plus is US$10 for US$20 of usage, Pro US$25 for US$50, Max US$100 for US$200 and Ultra US$250 for US$500. They renew automatically by card until disabled. Larger one-time top-ups advertise bonus credit, and failed requests are not charged. Promotional credit is service credit rather than withdrawable cash.

Data handling and acceptable use

The privacy notice says prompts and completions are not stored by default. The service retains an account email, hashed keys, billing records, request metadata, token counts, charges and SHA-256 fingerprints rather than the request text. Customers can enable seven-day request replay for a key. Third-party model hosts still process request and response content, and OpenAI's moderation API receives new messages and media prompts for safety classification.

Fewer refusals do not mean there are no rules. The acceptable-use policy prohibits illegal activity and several categories of harmful or non-consensual content. Model output may be inaccurate, offensive or unsuitable, and the terms place responsibility for prompts, outputs and end-user activity on the customer.

Our assessment

UncensoredLLM is most relevant to developers building fiction, roleplay, research or agent workflows that are poorly served by conventional refusal behaviour. Its familiar API surface, per-model pricing and observability tools make it more than a simple browser chatbot. The breadth of the catalogue is useful for experimentation, but model names that resemble well-known commercial products should not be assumed to be official versions from those vendors.

Before production use, teams should benchmark the exact model IDs they intend to call, verify tool calling and structured output, understand which host processes the data and test fallback behaviour. Buyers should also compare the total routed price and operational guarantees with direct hosting or self-deployment; the terms do not provide a service-level agreement.