Abliterated
Abliterated LLM API
Abliterated Kimi K3.1 Pro, Hy4 and DeepSeek V4.1 Flash 1017, plus 23 more text models, through the OpenAI and Anthropic APIs. Refusals are removed from the weights, not talked past.
No moderation layerPrompt logging off by defaultOpenAI and Anthropic clients
kimi-k3.1-pro-uncensoredBest at cyber work. Abliterated, 1M context.
- Input
- $5.00$4.00
- Cached
- $0.50
- Output
- $25.00$20.00
- Context
- 1M
hy4-uncensoredAbliterated, 1M context, cheap output.
- Input
- $1.60
- Cached
- $0.30
- Output
- $3.20
- Context
- 1M
deepseek-v4.1-flash-1017-uncensoredAbliterated DeepSeek, 1M context. The cheap one.
- Input
- $0.44$0.28
- Cached
- $0.07
- Output
- $1.50$1.00
- Context
- 1M
USD per 1M tokens. Struck prices are list rate; the live rate is what you pay.
Understand the model
Three words, three different things.
Uncensored and abliterated describe how a model was changed. Thinking describes how it answers. They are not interchangeable, and the difference decides which one you want.
Behaviour
Uncensored
Tuned or configured to answer requests a mainstream model declines. The model itself is unchanged at the weights.
Some GLM variants carry a built-in system prompt of roughly 14,000 tokens, billed as input on every request.
Model weights
Abliterated
The refusal behaviour is removed from the weights. There is no prompt to talk past, because nothing is layered on top.
GLM-5.3 Abliterated and GLM-5.3 Flash Abliterated were verified with native tool calls in Roo Code.
Reasoning
Thinking
Reasons before it answers. Reasoning tokens bill at the output rate, so a thinking model costs more per answer than its input price suggests.
Leave reasoning_effort unset to keep the default thinking mode.
Every model, its id and its rate
Text: USD per 1M tokens.
Text (26)
| Model | Model id | Use | Context | Input $/M | Cached $/M | Output $/M | Max output | Try |
|---|---|---|---|---|---|---|---|---|
| Kimi K3.1 ProChat · Kimi | kimi-k3.1-pro-uncensored | Best at cyber work. Abliterated, 1M context. | 1M | $5.00$4.00 | $0.50 | $25.00$20.00 | 944K | Chat |
| Hy4Chat · Hy4 | hy4-uncensored | Abliterated, 1M context, cheap output. | 1M | $1.60 | $0.30 | $3.20 | 66K | Chat |
| DeepSeek V4.1 Flash 1017Chat · DeepSeek | deepseek-v4.1-flash-1017-uncensored | Abliterated DeepSeek, 1M context. The cheap one. | 1M | $0.44$0.28 | $0.07 | $1.50$1.00 | 131K | Chat |
| GLM-5.3 Flash UncensoredChat · GLM | glm-5.3-flash-uncensored | Cheapest uncensored text model here. | — | $0.05 | $0.01 | $0.05 | — | Chat |
| GLM-5.3 UncensoredChat · GLM | glm-5.3-uncensored | Full GLM-5.3 with refusals relaxed. | — | $0.10 | $0.02 | $0.10 | — | Chat |
| GLM-5.3 Flash Cybersecurity UncensoredChat · GLM | glm-5.3-flash-cybersecurity-uncensored | Security and detection work. | — | $1.05 | $0.53 | $3.50 | — | Chat |
| Apex Flash 1 AbliteratedChat · GLM | glm-5.3-flash-apex-abliterated | Abliterated: refusals removed from the weights. | — | $0.30 | $0.15 | $1.00 | — | Chat |
| GLM-5.3 Flash AbliteratedChat · GLM | glm-5.3-flash-abliterated | Abliterated GLM-5.3 Flash. | — | $0.30 | $0.10 | $1.22 | — | Chat |
| GLM-5.3 AbliteratedChat · GLM | glm-5.3-abliterated | Abliterated GLM-5.3, native tool calls. | — | $1.40 | $0.40 | $1.40 | — | Chat |
| GPT OSS 120B UncensoredChat · GPT | gpt-oss-120b-uncensored | Open-weights GPT, uncensored. | — | $0.22 | $0.22 | $1.46 | — | Chat |
| DeepSeek V4 Pro UncensoredChat · DeepSeek | deepseek-v4-pro-uncensored | Full V4 Pro with refusals relaxed. | — | $0.50 | $0.10 | $0.50 | — | Chat |
| DeepSeek V4.1 Flash Uncensored Thinking 1017Chat · DeepSeek | deepseek-v4.1-flash-1017-uncensored-thinking | Same build, reasons before answering. | — | $0.66 | $0.02 | $2.64 | — | Chat |
| DeepSeek V4.1 Flash AbliteratedChat · DeepSeek | deepseek-v4.1-flash-abliterated | Abliterated: refusals removed from the weights. | — | $2.10 | $0.36 | $2.10 | — | Chat |
| DeepSeek V4.1 Flash Abliterated FastChat · DeepSeek | deepseek-v4.1-flash-abliterated-fast | Abliterated, faster route, lower rate. | — | $1.40 | $0.32 | $1.40 | — | Chat |
| MiMo V2.6 Flash AbliteratedChat · MiMo | mimo-v2.6-flash-abliterated | Abliterated MiMo Flash. | — | $1.00 | $0.55 | $3.00 | — | Chat |
| MiMo V2.6 Flash UncensoredChat · MiMo | mimo-v2.6-flash-uncensored | Uncensored MiMo Flash. | — | $1.00 | $0.40 | $3.00 | — | Chat |
| MiMo V2.6 Flash Uncensored ThinkingChat · MiMo | mimo-v2.6-flash-uncensored-thinking | Uncensored MiMo Flash, reasons first. | — | $1.00 | $0.40 | $3.00 | — | Chat |
| MiMo V2.6 Pro UncensoredChat · MiMo | mimo-v2.6-pro-uncensored | The larger MiMo, uncensored. | — | $2.29 | $0.02 | $4.57 | — | Chat |
| MiniMax M3 UncensoredChat · MiniMax | minimax-m3-uncensored | Uncensored MiniMax M3. | — | $1.50 | $0.38 | $8.00 | — | Chat |
| Nemotron 3 Ultra 550B UncensoredChat · Nemotron | nemotron-3-ultra-550b-a55b-uncensored | NVIDIA 550B, uncensored. | — | $1.69 | $0.43 | $9.90 | — | Chat |
| Hermes 4 405BChat · Hermes | hermes-4-405b-uncensored | Hermes 4 405B, uncensored. | — | $1.80 | $1.80 | $7.20 | — | Chat |
| Qwen 3.8 27B AbliteratedChat · Qwen | qwen3.8-27b-abliterated | Abliterated Qwen 3.8 27B. | — | $0.51 | $0.07 | $2.27 | — | Chat |
| Outrun Red Team FlashChat · Outrun | outrun-redteam-flash | Tuned for red-team prompts. | — | $1.35 | $0.24 | $4.50 | — | Chat |
| Outrun Red Team ProChat · Outrun | outrun-redteam-pro | The larger red-team model. | — | $2.25 | $0.64 | $9.00 | — | Chat |
| GLM-5.3 Flash Code UncensoredCode · GLM | glm-5.3-flash-code-uncensored | Uncensored coding model. The expensive one. | — | $12.00 | $2.40 | $12.00 | — | Chat |
| Uncens Code Gemma 4 27BCode · Gemma · ours | uncens-code-gemma-4-27b | Uncensored Gemma coder. | — | Free | — | Free | — | Chat |
- Kimi K3.1 Pro$4.00 / $20.00
Chat · Kimi
Best at cyber work. Abliterated, 1M context.
kimi-k3.1-pro-uncensored - Hy4$1.60 / $3.20
Chat · Hy4
Abliterated, 1M context, cheap output.
hy4-uncensored - DeepSeek V4.1 Flash 1017$0.28 / $1.00
Chat · DeepSeek
Abliterated DeepSeek, 1M context. The cheap one.
deepseek-v4.1-flash-1017-uncensored - GLM-5.3 Flash Uncensored$0.05 / $0.05
Chat · GLM
Cheapest uncensored text model here.
glm-5.3-flash-uncensored - GLM-5.3 Uncensored$0.10 / $0.10
Chat · GLM
Full GLM-5.3 with refusals relaxed.
glm-5.3-uncensored - GLM-5.3 Flash Cybersecurity Uncensored$1.05 / $3.50
Chat · GLM
Security and detection work.
glm-5.3-flash-cybersecurity-uncensored - Apex Flash 1 Abliterated$0.30 / $1.00
Chat · GLM
Abliterated: refusals removed from the weights.
glm-5.3-flash-apex-abliterated - GLM-5.3 Flash Abliterated$0.30 / $1.22
Chat · GLM
Abliterated GLM-5.3 Flash.
glm-5.3-flash-abliterated - GLM-5.3 Abliterated$1.40 / $1.40
Chat · GLM
Abliterated GLM-5.3, native tool calls.
glm-5.3-abliterated - GPT OSS 120B Uncensored$0.22 / $1.46
Chat · GPT
Open-weights GPT, uncensored.
gpt-oss-120b-uncensored - DeepSeek V4 Pro Uncensored$0.50 / $0.50
Chat · DeepSeek
Full V4 Pro with refusals relaxed.
deepseek-v4-pro-uncensored - DeepSeek V4.1 Flash Uncensored Thinking 1017$0.66 / $2.64
Chat · DeepSeek
Same build, reasons before answering.
deepseek-v4.1-flash-1017-uncensored-thinking - DeepSeek V4.1 Flash Abliterated$2.10 / $2.10
Chat · DeepSeek
Abliterated: refusals removed from the weights.
deepseek-v4.1-flash-abliterated - DeepSeek V4.1 Flash Abliterated Fast$1.40 / $1.40
Chat · DeepSeek
Abliterated, faster route, lower rate.
deepseek-v4.1-flash-abliterated-fast - MiMo V2.6 Flash Abliterated$1.00 / $3.00
Chat · MiMo
Abliterated MiMo Flash.
mimo-v2.6-flash-abliterated - MiMo V2.6 Flash Uncensored$1.00 / $3.00
Chat · MiMo
Uncensored MiMo Flash.
mimo-v2.6-flash-uncensored - MiMo V2.6 Flash Uncensored Thinking$1.00 / $3.00
Chat · MiMo
Uncensored MiMo Flash, reasons first.
mimo-v2.6-flash-uncensored-thinking - MiMo V2.6 Pro Uncensored$2.29 / $4.57
Chat · MiMo
The larger MiMo, uncensored.
mimo-v2.6-pro-uncensored - MiniMax M3 Uncensored$1.50 / $8.00
Chat · MiniMax
Uncensored MiniMax M3.
minimax-m3-uncensored - Nemotron 3 Ultra 550B Uncensored$1.69 / $9.90
Chat · Nemotron
NVIDIA 550B, uncensored.
nemotron-3-ultra-550b-a55b-uncensored - Hermes 4 405B$1.80 / $7.20
Chat · Hermes
Hermes 4 405B, uncensored.
hermes-4-405b-uncensored - Qwen 3.8 27B Abliterated$0.51 / $2.27
Chat · Qwen
Abliterated Qwen 3.8 27B.
qwen3.8-27b-abliterated - Outrun Red Team Flash$1.35 / $4.50
Chat · Outrun
Tuned for red-team prompts.
outrun-redteam-flash - Outrun Red Team Pro$2.25 / $9.00
Chat · Outrun
The larger red-team model.
outrun-redteam-pro - GLM-5.3 Flash Code Uncensored$12.00 / $12.00
Code · GLM
Uncensored coding model. The expensive one.
glm-5.3-flash-code-uncensored - Uncens Code Gemma 4 27BFree
Code · Gemma · ours
Uncensored Gemma coder.
uncens-code-gemma-4-27b
Image (12)
| Model | Model id | Use | Context | Input $/M | Cached $/M | Output $/M | Max output | Try |
|---|---|---|---|---|---|---|---|---|
| Anime Image UncensoredImage · SDXL | anime-image-uncensored | Anime and illustration. | — | — | — | $0.06/img | — | Chat |
| CyberRealistic XL UncensoredImage · SDXL | cyberrealistic-xl-uncensored | Photoreal SDXL. | — | — | — | $0.06/img | — | Chat |
| WAI Illustrious SDXL UncensoredImage · SDXL | illustrious-xl-uncensored | Illustrious SDXL, anime-leaning. | — | — | — | $0.06/img | — | Chat |
| Lustify SDXL UncensoredImage · SDXL | lustify-sdxl-uncensored | Photo-style SDXL, no filter. | — | — | — | $0.07/img | — | Chat |
| Seedream 5.0 Lite Edit UncensoredImage · Seedream | seedream-5.0-lite-edit-uncensored | Image edit. Needs a reference image. | — | — | — | $0.07/img | — | Chat |
| Qwen Image 2.1 UncensoredImage · Qwen | qwen-image-2.1-uncensored | Qwen image, 1K and 2K. | — | — | — | $0.08/img | — | Chat |
| Vision Image UncensoredImage · Vision | vision-image-uncensored | General-purpose image. | — | — | — | $0.08/img | — | Chat |
| Chroma UncensoredImage · Chroma | chroma-uncensored | Chroma, no content filter. | — | — | — | $0.16/img | — | Chat |
| Qwen Image 3 UncensoredImage · Qwen | qwen-image-3-uncensored | Newer Qwen image model. | — | — | — | $0.16/img | — | Chat |
| Qwen Edit UncensoredImage · Qwen | qwen-edit-uncensored | Instruction-based image edit. | — | — | — | $0.16/img | — | Chat |
| Vision Pro Image UncensoredImage · Vision | vision-pro-image-uncensored | Higher-detail image, slower. | — | — | — | $0.19/img | — | Chat |
| HunyuanImage 3.0 Instruct UncensoredImage · Hunyuan | hunyuan-image-3-instruct-uncensored | Tencent Hunyuan, instruction-driven. | — | — | — | $0.23/img | — | Chat |
- Anime Image Uncensored$0.06/img
Image · SDXL
Anime and illustration.
anime-image-uncensored - CyberRealistic XL Uncensored$0.06/img
Image · SDXL
Photoreal SDXL.
cyberrealistic-xl-uncensored - WAI Illustrious SDXL Uncensored$0.06/img
Image · SDXL
Illustrious SDXL, anime-leaning.
illustrious-xl-uncensored - Lustify SDXL Uncensored$0.07/img
Image · SDXL
Photo-style SDXL, no filter.
lustify-sdxl-uncensored - Seedream 5.0 Lite Edit Uncensored$0.07/img
Image · Seedream
Image edit. Needs a reference image.
seedream-5.0-lite-edit-uncensored - Qwen Image 2.1 Uncensored$0.08/img
Image · Qwen
Qwen image, 1K and 2K.
qwen-image-2.1-uncensored - Vision Image Uncensored$0.08/img
Image · Vision
General-purpose image.
vision-image-uncensored - Chroma Uncensored$0.16/img
Image · Chroma
Chroma, no content filter.
chroma-uncensored - Qwen Image 3 Uncensored$0.16/img
Image · Qwen
Newer Qwen image model.
qwen-image-3-uncensored - Qwen Edit Uncensored$0.16/img
Image · Qwen
Instruction-based image edit.
qwen-edit-uncensored - Vision Pro Image Uncensored$0.19/img
Image · Vision
Higher-detail image, slower.
vision-pro-image-uncensored - HunyuanImage 3.0 Instruct Uncensored$0.23/img
Image · Hunyuan
Tencent Hunyuan, instruction-driven.
hunyuan-image-3-instruct-uncensored
Video (3)
| Model | Model id | Use | Context | Input $/M | Cached $/M | Output $/M | Max output | Try |
|---|---|---|---|---|---|---|---|---|
| LTX-2.3 Image-to-Video UncensoredVideo · LTX | ltx-2.3-i2v-uncensored | Image to video, 480p. | — | — | — | $0.16/sec | — | Chat |
| Wan 3.0 Video UncensoredVideo · Wan | wan-3.0-video-uncensored | Text to video, 480p. | — | — | — | $0.40/sec | — | Chat |
| Wan 3.0 Image-to-Video UncensoredVideo · Wan | wan-3.0-i2v-uncensored | Image to video, 480p. | — | — | — | $0.40/sec | — | Chat |
- LTX-2.3 Image-to-Video Uncensored$0.16/sec
Video · LTX
Image to video, 480p.
ltx-2.3-i2v-uncensored - Wan 3.0 Video Uncensored$0.40/sec
Video · Wan
Text to video, 480p.
wan-3.0-video-uncensored - Wan 3.0 Image-to-Video Uncensored$0.40/sec
Video · Wan
Image to video, 480p.
wan-3.0-i2v-uncensored
Rows marked ours run on our own upstream and answer today. The remaining 40 come through our partnership with outrun and are being connected to this gateway; until that channel is live they return an error rather than an answer. Ids ending in :free share one allowance of 100 requests a day across every free model, in Chat and through the API.
Connect
The same call as every other model.
One base URL, one key, the OpenAI format. Switching to an uncensored model is one string in the request.
- Base URL
- https://api.oneaiapikey.com/v1
- Header
- Authorization: Bearer sk-…
curl https://api.oneaiapikey.com/v1/chat/completions \
-H "Authorization: Bearer $ONE_AI_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "glm-5.3-flash-uncensored",
"messages": [{ "role": "user", "content": "Your prompt here" }]
}'from openai import OpenAI
client = OpenAI(
base_url="https://api.oneaiapikey.com/v1",
api_key="sk-...",
)
r = client.chat.completions.create(
model="glm-5.3-flash-uncensored",
messages=[{"role": "user", "content": "Your prompt here"}],
)
print(r.choices[0].message.content)import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.oneaiapikey.com/v1",
apiKey: process.env.ONE_AI_KEY,
});
const r = await client.chat.completions.create({
model: "glm-5.3-flash-uncensored",
messages: [{ role: "user", content: "Your prompt here" }],
});
console.log(r.choices[0].message.content);Swap glm-5.3-flash-uncensored for any id in the table above. Image and video models take the same key on their own endpoints.
Handling
What happens to your prompts.
- Not written to our logs
- By default the content of your prompts and responses is relayed in memory to complete the request and is not written to our usage logs. Only metadata is stored: model, time, token counts, status and charge.
- Prompt logging is opt-in
- It is off by default and applies only to a key you switch it on for. Turning it off deletes that key's rows at once.
- Not used for training
- We never use prompts, files or outputs that pass through the API to train or fine-tune a model, and we do not sell them.
- Upstream providers see the request
- To produce an answer your request goes to the provider running that model, under its own terms. That is true of every gateway, ours included. Privacy Policy
Questions
The ones that come up first.
What is the difference between uncensored and abliterated?
Uncensored is behaviour: the model is tuned or configured to answer what a mainstream model declines. Abliterated is the weights: the refusal direction is removed from the model itself. Abliterated is the more thorough of the two, and it is what the security and detection work here runs on.
Which model should I start with?
GLM-5.3 Flash Uncensored is the cheapest text model at $0.05 in and $0.05 out per 1M tokens.
How much does the built-in system prompt add per request?
Some GLM uncensored variants ship with a system prompt of about 14,000 tokens. It is billed as input on every request, so a short prompt still pays for it. At $0.05 per 1M input tokens that is well under a cent, but it is why a GLM uncensored call costs more than its prompt looks.
Do you store my prompts and responses?
No. By default the content of your prompts and responses is relayed in memory to complete the request and is not written to our usage logs; only metadata is kept. Prompt logging exists, but it is off by default, applies only to a key you switch it on for, and is deleted when you switch it off. The Privacy Policy is the binding text.
Does a lower refusal rate change what I am allowed to do?
No. The Acceptable Use Policy covers every model here exactly as it covers the rest of the catalogue. Child safety rules apply without exception. Fewer refusals from the model does not change what the terms allow.
How do I pay?
Prepaid credit in US dollars, from $1.00. No subscription, no per-seat fee, and the balance is shared with every other model on the gateway.
Can I use these from a coding agent or a chat app?
Yes. Anything that accepts an OpenAI-compatible base URL and key works: Roo Code, Cline, SillyTavern, LibreChat, the OpenAI SDKs. Point it at the base URL below and paste the model id from the table.
You are still the one accountable.
Running an uncensored model does not change what you may lawfully do with the output. The Acceptable Use Policy applies here exactly as it does to every other model. What changes is that the model will not stop you at the prompt.
Same balance, same endpoint, same key. New accounts get $10 of credit to test with.