hide.st
Hide Router

Uncensored AI for your coding agent.

36 uncensored and abliterated models that call tools, so they work in OpenCode, Claude Code, Cline and any OpenAI-compatible agent. They're built on GLM 5.3, DeepSeek V4.1, Qwen 3.8 and others, and refuse less often on legitimate work such as security research. One key, paid per request with crypto.

Get an API keySee the setup
An uncensored model in your coding agent
  1. Point your agent at hide.st: one base URL and one key.
  2. Pick any of 36 uncensored models that can call tools.
  3. The agent works as usual: it reads, searches and runs your code.
  4. Every request reports exactly what it cost.
  5. It's paid from your crypto balance. No subscription, no email.

Where to start

Every model here calls tools and takes a long context. Copy an ID into your agent's config, or open the model for its full details.

  • Hardest tasks

    Abliterated Model Large V2

    Built on the full GLM-5.3 and modified in its weights to refuse less.

    abliteration-ai/abliterated-model-large-v21M context$3.3 in, $5.5 out

  • Same base, lower price

    GLM 5.3 Uncensored

    An uncensored GLM 5.3. It's described as experimental, so quality can vary.

    z-ai/glm-5.3-uncensored1M context$1.38 in, $2.48 out

  • Cheapest with 1M context

    DeepSeek V4.1 Flash Uncensored

    A lower-refusal DeepSeek V4.1 Flash that also reads images.

    deepseek/deepseek-v4.1-flash-uncensored1M context$0.363 in, $1.45 out

  • Authorized security work

    OrcaCyber Zero 1.5

    Post-trained for vulnerability research and other security work you have permission for.

    orca/orcacyber-zero-1.5999K context$3.3 in, $8.25 out

Set it up

Point your tool at https://hide.st/api/v1, or https://hide.st/api for Claude Code, and use a hide.st key.

  • OpenCode
  • Claude Code
  • Cline
  • Kilo Code
  • Aider
  • Open WebUI
  • LibreChat
  • SillyTavern
Setup guides for each tool
{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "hide": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Hide Router",
      "options": {
        "baseURL": "https://hide.st/api/v1",
        "apiKey": "{env:HIDE_API_KEY}"
      },
      "models": {
        "abliteration-ai/abliterated-model-large-v2": { "name": "Abliterated Model Large V2" },
        "z-ai/glm-5.3-uncensored": { "name": "GLM 5.3 Uncensored" }
      }
    }
  },
  "model": "hide/abliteration-ai/abliterated-model-large-v2"
}

Every uncensored model for agents

36 models with tool calling. Prices per million tokens, including our margin.

Compare them in the model list

One key for every model

Switch between uncensored models and Claude, GPT or Gemini per task, without another account or another API key.

A spending limit per agent

Give each tool its own key with a cap, so an agent stuck in a loop can't spend more than you allowed.

Your code isn't kept

Requests pass straight to the company that runs the model and aren't stored. We keep the model, token counts and cost.

No account details

Your account is a random key, not an email. Top up with crypto from $1 and pay per request, with no subscription.

Fewer refusals, same rules

These models decline less often, but that isn't permission. Our acceptable use rules apply to every model: don't write malware, and only test systems you own or have permission to test.

Questions about uncensored coding models

What is an uncensored coding model?

A model that writes code and calls tools like any other, but has been modified or retrained to refuse less often. Abliterated models have the refusal behaviour removed from their weights; fine-tuned ones were retrained on data with fewer refusals. How well each one codes depends on the model it's built on.

Which uncensored model is best for coding?

Abliterated Model Large V2 for the hardest tasks, GLM 5.3 Uncensored for the same base at a lower price, DeepSeek V4.1 Flash Uncensored as the cheapest, with a 1M-token context and OrcaCyber Zero 1.5 for authorized security work. All of them call tools, so they work in coding agents.

Can I use uncensored models in Claude Code?

Yes. hide.st speaks the Anthropic Messages API, so Claude Code runs on any model here: set ANTHROPIC_BASE_URL to https://hide.st/api, your hide.st key as ANTHROPIC_AUTH_TOKEN and the model ID as ANTHROPIC_MODEL.

Do they work with OpenCode, Cline and other agents?

Yes. Use an OpenAI-compatible provider with the base URL https://hide.st/api/v1, a hide.st key and the model ID. Tool calls, streaming and reasoning work as they do with any other model.

Do you store my code or prompts?

No. Requests pass straight to the company that runs the model and aren't stored. We keep only what billing needs: the model, token counts and cost.

How do I pay?

Top up with Bitcoin, Monero, USDT or other coins from $1. Each request is charged from your balance at the model's price plus our margin, with no subscription, and the balance doesn't expire.

Can I use them for security research?

Yes, for authorized work: your own systems, CTFs and engagements you have permission for. Our acceptable use rules forbid writing malware and testing systems without permission, whatever the model.