Pre-launch. You can create an account today; the models are switched on soon and nothing is charged until then.

Open-weight models, refusals removed

Uncensored AI for code and chat.

Abliterated open models behind an OpenAI-compatible API and a private chat. It answers the question you asked, without the lecture, and your prompts are not stored.

Get API key Open chat

Trial credit on sign-up. No card needed to try it.

Fig. 1Illustrative

Abliteration takes one direction, r, out of the weights. Everything else stays where it was.

Drop-in API

The OpenAI chat completions format, with streaming and tool calls. Change the base URL and the model name, keep your code.

Prompts are not stored

The gateway keeps token counts for billing and nothing else. Chat history lives in your browser, not on our servers.

Pay for what you use

Per-token prices on every model, prepaid credits, no minimum. A subscription only if you want flat-rate chat.

The same question, answered

An ordinary request that a filtered assistant often declines, and what an abliterated model does with it. The exchange is an example written for this page, not a recorded session.

Qwen 3.8 27B abliteratedIllustrative

You

I forgot the password of my own ZIP archive. Write a script that tries every line of a wordlist.

A filtered assistant

I can't help with bypassing password protection.

This model

import zipfile, sys

archive, wordlist = sys.argv[1], sys.argv[2]
z = zipfile.ZipFile(archive)
for line in open(wordlist, "rb"):
    pw = line.strip()
    try:
        z.extractall(pwd=pw)
        print("password:", pw.decode())
        break
    except (RuntimeError, zipfile.BadZipFile):
        continue

Models

Open-weight models with the refusal behaviour removed (abliteration), served from our own inference stack. Only permissively licensed models are listed.

Model	API id	Good for	Context	Input / 1M	Output / 1M
Qwen 3.8 27B abliteratedQwen, Apache-2.0	qwen3.8-27b-abliterated	General chat, coding, long context	196,000	$0.30	$1.20
Coding model, abliteratedPlanned, not available yet	n/a	Agentic coding with tool calls	128,000	n/a	n/a

Prices in US dollars per million tokens. Context in tokens.

One endpoint, the format you already use

Point any OpenAI client, editor or coding agent at AblitLLM. Streaming, tool calls and JSON output behave the way your code expects.

Usage and cost per request in your dashboard.
Keys you can name, rotate and revoke.
Errors in the OpenAI format, with clear types.

Read the API reference

curl https://ablitllm.com/v1/chat/completions \
  -H "Authorization: Bearer ab_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-27b-abliterated",
    "stream": true,
    "messages": [{"role": "user", "content": "Write a port scanner in Go."}]
  }'

What uncensored means here

The models have no refusal training left and we add no moral filter of our own. There is exactly one thing the service blocks, and it is stated here instead of hidden in the terms.

It answers

Security research: exploits, malware analysis, offensive tooling for systems you are allowed to test.
Blunt medical, legal and financial information without the disclaimers stacked on top.
Dark, violent or adult fiction between adults.
Profanity, insults, controversial opinions, and questions other assistants deflect.

It blocks

Sexual content involving minors. It is a crime to produce or host, so requests for it are rejected before they reach a model.
Nothing else is filtered. What you do with an answer is your responsibility, and using the service to break the law is against the acceptable use policy.

Pricing

Free

Trial credit on sign-up, chat and API.

$0.10 of credit when you sign up
Chat and API on every live model
Usage dashboard and API keys

Create account

Pro

$15per month

Chat on the included models without spending credits, within a daily fair-use limit.

Up to 500 chat messages a day
Everything in Free
API billed per token from your credits

Choose Pro

Pay as you go

$0.30in, $1.20 out, per 1M tokens

Prepaid credits billed per token. No subscription, no minimum.

Credit packs of $5, $20, $50
Credits do not expire
No subscription and no minimum spend

Get API key

Questions

What is an abliterated model?+

An open-weight model whose internal refusal direction has been identified and removed. The model keeps its knowledge and skills; what goes away is the trained habit of declining and lecturing.

Is this legal?+

The models are published under permissive open licences that allow modification and commercial use. What the law cares about is what you do with the output, as with any other tool, and that responsibility is yours.

Do you log my prompts?+

No. For each request we store the model, the token counts, the cost and the time, because that is what billing needs. The text of prompts and answers is not written to disk. Chat conversations are saved in your browser only.

Can I use it with my editor or coding agent?+

Yes, if the tool lets you set an OpenAI-compatible base URL and API key. Streaming and tool calls are supported.

Which models do you serve?+

Today: Qwen 3.8 27B abliterated. The models table above is the live list; planned models are marked as such and cannot be called yet.

Ask it the thing the other one refused.

Get API key Open chat