Eligible new accounts get $0.10 in free API credit. Once per account; abuse limits apply. Create an account

Outrun
Sign in Get your key

One AI API with uncensored and abliterated models.

Outrun runs 25 uncensored and abliterated LLMs, 7 uncensored image models and an uncensored video model through one OpenAI-compatible API key. Text starts at $0.10 per 1M tokens, billed from prepaid USD credit with no subscription.

Try uncensored models Compare prices

$0.10free credit for eligible new accounts, and 10 free web searches a day in the Playground.

25uncensored and abliterated LLMs
7uncensored image models
$0.10lowest input price per 1M tokens
$0.40per second of 480p video
  • Uncensored$0.10 in / $0.10 out
    GLM-5.3 Flash Uncensoredglm-5.3-flash-uncensored

    Lowest-priced uncensored LLM on Outrun

  • Abliterated$1.40 in / $1.40 out
    GLM-5.3 Abliteratedglm-5.3-abliterated

    Native tool calls verified in Roo Code

  • Uncensored$0.22 in / $1.46 out
    GPT OSS 120B Uncensoredgpt-oss-120b-uncensored

    Uncensored build of the open-weight GPT OSS 120B

What is the difference between uncensored and abliterated models?

Uncensored models are fine-tuned or configured to refuse far less than mainstream chat models. Abliterated models have their refusal behaviour removed from the model weights. Both run through the same API, so you can compare them by changing one model ID.

  • Uncensored

    Tuned or configured to answer requests that mainstream models decline. Some GLM variants use a built-in system prompt of about 14,000 tokens, which is billed as input on every request.

  • Abliterated

    Refusal behaviour is removed from the weights. GLM-5.3 Abliterated and GLM-5.3 Flash Abliterated were verified with native tool calls in Roo Code.

  • Thinking

    MiMo V2.6 Flash Uncensored Thinking reasons before it answers, and its reasoning tokens use the output rate. Leave reasoning_effort unset so it keeps its default thinking mode.

  • Alternative routes

    A model marked (2) or (3) is an alternative route to the same base model, with its own ID and price. Use the exact ID of the version you want.

Uncensored LLM API prices

The lowest-priced uncensored LLM on Outrun is GLM-5.3 Flash Uncensored at $0.10 per 1M input tokens and $0.10 per 1M output tokens. All 25 uncensored and abliterated text models are listed below, grouped by base model and sorted by price, from $0.10 to $14.68 per 1M output tokens.

Text modelsUSD per 1M tokens
ModelInputOutputCached input
GLM-5.3 FlashGuide
GLM-5.3 Flash Uncensoredglm-5.3-flash-uncensoredBuilt-in system prompt of about 14k tokens, billed as input$0.10$0.10$0.01
GLM-5.3 Flash Uncensored (2)glm-5.3-flash-uncensored-2Alternative route$0.17$0.17$0.02
Apex Flash 1 Abliteratedglm-5.3-flash-apex-abliteratedRefusals removed in the model weights, not by a prompt$0.30$1.00$0.15
GLM-5.3 Flash Abliteratedglm-5.3-flash-abliteratedRefusals removed in the model weights, not by a prompt · Native tool calls verified in Roo Code$0.40$0.40$0.10
GLM-5.3 Flash Uncensored (3)glm-5.3-flash-uncensored-3Alternative route$0.40$1.60$0.14
GLM-5.3 Flash Abliterated (2)glm-5.3-flash-abliterated-2Refusals removed in the model weights, not by a prompt · alternative route$1.32$1.32$0.36
GLM-5.3 Flash Code Uncensoredglm-5.3-flash-code-uncensoredUncensored version of GLM-5.3 Flash Code$12.00$12.00$12.00
GLM-5.3Guide
GLM-5.3 Uncensoredglm-5.3-uncensoredBuilt-in system prompt of about 14k tokens, billed as input$0.20$0.20$0.06
GLM-5.3 Uncensored (2)glm-5.3-uncensored-2Alternative route$0.35$0.35$0.10
GLM-5.3 Abliteratedglm-5.3-abliteratedRefusals removed in the model weights, not by a prompt · Native tool calls verified in Roo Code$1.40$1.40$0.40
GLM-5.3 Uncensored (3)glm-5.3-uncensored-3Alternative route$2.50$4.50$0.60
GLM-5.3 Abliterated (2)glm-5.3-abliterated-2Refusals removed in the model weights, not by a prompt · alternative route$4.50$4.50$1.32
GPT OSS
GPT OSS 120B Uncensoredgpt-oss-120b-uncensoredUncensored version of GPT OSS 120B$0.22$1.46$0.22
DeepSeekGuide
DeepSeek V4 Pro Uncensoreddeepseek-v4-pro-uncensoredUncensored version of DeepSeek V4 Pro$0.50$0.50$0.10
DeepSeek V4.1 Flash Abliterated (2)deepseek-v4.1-flash-abliterated-2Refusals removed in the model weights, not by a prompt · alternative route$0.70$0.70$0.12
DeepSeek V4.1 Flash Abliterateddeepseek-v4.1-flash-abliteratedRefusals removed in the model weights, not by a prompt$2.10$2.10$0.36
MiMo
MiMo V2.6 Flash Abliteratedmimo-v2.6-flash-abliteratedRefusals removed in the model weights, not by a prompt$1.00$3.00$0.55
MiMo V2.6 Flash Uncensoredmimo-v2.6-flash-uncensoredUncensored version of MiMo V2.6 Flash$1.00$3.00$0.40
MiMo V2.6 Flash Uncensored Thinkingmimo-v2.6-flash-uncensored-thinkingReasoning tokens use the output rate$1.00$3.00$0.40
Outrun Red Team
Outrun Red Team Flashoutrun-redteam-flashUncensored text model$1.35$4.50$0.24
Outrun Red Team Prooutrun-redteam-proUncensored text model$2.25$9.00$0.64
MiniMax
MiniMax M3 Uncensoredminimax-m3-uncensoredUncensored version of MiniMax M3$1.50$8.00$0.38
Nemotron
Nemotron 3 Ultra 550B Uncensorednemotron-3-ultra-550b-a55b-uncensoredUncensored version of Nemotron 3 Ultra 550B$1.69$9.90$0.43
Hermes
Hermes 4 405Bhermes-4-405b-uncensoredUncensored text model$1.80$7.20$1.80
Kimi
Kimi K2.6 Uncensoredkimi-k2.6-uncensoredUncensored version of Kimi K2.6$2.34$14.68$0.59

Rates are shown rounded up to the cent; billing uses the exact rate. The cached rate applies when a request reuses a cached prompt prefix. MiMo models have a 1,048,576-token context and a 65,536-token maximum output. Context window and maximum output are separate limits, so set each one on its own in your app. The GLM-5.3 and DeepSeek guides compare each version of those models.

Uncensored image and video generation API

Outrun runs 7 uncensored image models, from under one cent per image, and an uncensored video model from $0.40 per second of output. Images use POST /v1/images/generations and video uses /v1/videos, with the same key and USD balance as text models.

Image modelsUSD per image
ModelPer image
Chroma Uncensoredchroma-uncensored$0.01
Vision Image Uncensoredvision-image-uncensored$0.01
Lustify SDXL Uncensoredlustify-sdxl-uncensored$0.02
Anime Image Uncensoredanime-image-uncensored$0.04
Seedream 5.0 Lite Edit Uncensoredseedream-5.0-lite-edit-uncensoredNeeds a reference image$0.07
Qwen Image 2.1 Uncensoredqwen-image-2.1-uncensored1K$0.08
Qwen Image 2.1 Uncensored, 2K$0.40
Vision Pro Image Uncensoredvision-pro-image-uncensored$0.09
Video modelsUSD per second of output
ModelPer second
Wan 3.0 Video Uncensoredwan-3.0-video-uncensored480p$0.40
Wan 3.0 Video Uncensored, 720p$0.80
Wan 3.0 Video Uncensored, 1080p$1.60
  • First image. Start with vision-image-uncensored for text-to-image.
  • Reference images. Seedream 5.0 Lite Edit Uncensored needs a reference image, and Qwen Image 2.1 Uncensored accepts input.image_urls with a 1K or 2K input.resolution.
  • Video. Send a prompt with seconds from 2 to 30 and a resolution of 480p, 720p or 1080p.
No setupTry any of these in the PlaygroundGenerate images and video or chat with any uncensored model in your browser, with the same key and prices as the API.

How do I connect an app to uncensored models?

  1. 01

    Get your key

    Open the Telegram bot and copy your personal key. Paste it only into your app's private key field.

  2. 02

    Add an OpenAI-compatible provider

    Set the base URL to https://api.outrun.to/v1 exactly as shown, without /chat/completions.

  3. 03

    Pick a model ID

    Start with glm-5.3-abliterated, the model the setup guides test with, and save other IDs as separate profiles to compare them.

  4. 04

    Send a short test

    Ask for "Reply with OK only." A short reply confirms the connection, and the test is billed from your credit like any other request.

Why the guide uses Roo Code. Coding agents send their own built-in system prompt with every request, and in many of them that prompt tells the model to refuse certain tasks, which can override an uncensored model. Roo Code is the agent Outrun tested these models with, including native tool calls on GLM-5.3 Abliterated and GLM-5.3 Flash Abliterated.

POST/v1/chat/completions 
curl https://api.outrun.to/v1/chat/completions \
  -H "Authorization: Bearer $OUTRUN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3-abliterated",
    "messages": [
      {"role": "user", "content": "Hello!"}
    ]
  }'
Swap the model
GLM-5.3 Abliterated$1.40 in/$1.40 outper 1M tokens All uncensored prices

Keep your key in an environment variable, never in client-side code.

Setup guides

No computer neededSet it up on your phoneA 9-step guide with a screenshot for every step connects Cumbersome on iPhone to your Outrun key. AnythingLLM on Android uses the same values.

Does Outrun store my prompts and responses?

No. Outrun runs zero data retention on every model, uncensored and abliterated ones included: prompts, responses and uploads stay out of its request logs, and API calls are never added to your chat history. For each request Outrun keeps only the billing record: the model, request ID, time, token counts, status and charge.

Read the full privacy policy

POST /v1/chat/completionsExample request record
Model
glm-5.3-abliterated
Tokens
1,284 in · 312 out
Charge
$0.0023 at $1.40 in / $1.40 out per 1M
Status
200 OK
Prompt
Not stored
Response
Not stored

The billing record is all Outrun keeps for an API request.

  • No content in request logs

    Outrun's no-content-logging standard applies to every model. Prompts, responses, uploads and raw provider error bodies are kept out of operational request logs.

  • API calls leave no transcript

    Direct API text calls are not added to account chat history. Apps you connect, such as Roo Code or a mobile client, can keep their own copies on your device.

  • Playground history is your choice

    Playground text chats start temporary and are gone when you close the page. Turn on saved history and conversations are encrypted at rest, and you can delete them at any time. Image and video jobs keep their prompt and results so you can download them.

  • Provider ZDR on top

    Outrun's zero retention covers every model. Models that answer through an AI provider are also processed under that provider's retention policy, and DeepSeek V4.1 Flash 1017 carries a Provider ZDR badge, which adds zero retention at the model provider as well. Provider audit exceptions still apply, including records of generated text for up to 30 days.

What is not allowed on uncensored models?

The same acceptable use policy covers every Outrun model, uncensored and abliterated ones included. Fewer refusals from a model does not change what the terms allow.

  • Child safety

    Child sexual abuse material and sexual content involving anyone under 18 are prohibited across all Outrun models, including uncensored, abliterated, image and video models.

Uncensored API questions

The Roo Code guide covers model settings and error codes step by step.

How much does the built-in system prompt add per request?

GLM-5.3 Flash Uncensored and GLM-5.3 Uncensored bill about 14,000 extra input tokens on every request, which adds about $0.0014 and $0.0028 per request at their rates. For many short requests, compare the total with an abliterated model before you choose.

What settings do abliterated GLM models need in a coding agent?

Set the context window to 1,000,000 tokens and the maximum output to 65,536 tokens. With those limits, GLM-5.3 Abliterated and GLM-5.3 Flash Abliterated were verified with native tool calls in Roo Code, and the Roo Code guide shows each setting.

Try an uncensored model in the Playground.

Test any model in your browser with your own key and credit, then connect Roo Code or a mobile app with the same key.

Try the Playground Get your API key