# ABLITERATED.cloud

> Publicly downloadable Hugging Face model weights labeled uncensored and abliterated, available through one token-protected OpenAI-compatible API powered by vLLM and scale-to-zero Modal GPUs.

ABLITERATED.cloud is a private evaluation service for four exact model profiles. Expensive hardware profiles require explicit operator approval and a confirmed cloud budget before deployment.

## Primary resources

- [Human landing page](https://abliterated.cloud/): Product overview, models, costs, security, and access.
- [Markdown landing page](https://abliterated.cloud/index.md): Compact agent-friendly product documentation.
- [Full agent context](https://abliterated.cloud/llms-full.txt): Expanded operational and model details.
- [OpenAPI catalog](https://abliterated.cloud/openapi.json): Current API contract and model identifiers.
- [Authentication guide](https://abliterated.cloud/auth.md): Bearer-token and lifecycle requirements.
- [Agent skill](https://abliterated.cloud/skills/abliterated-cloud/SKILL.md): Safe instructions for agents using the API.
- [Source repository](https://github.com/eminogrande/mn-uncensored): Open-source implementation, tests, operations, release notes, and incident report.
- [Exact model record](https://github.com/eminogrande/mn-uncensored/blob/main/docs/MODELS.md): Hugging Face artifacts, revisions, licenses, context, parsers, and hardware.

## Model field notes

- [Huihui-Qwen3.6 35B A3B](https://abliterated.cloud/blog/qwen3-6-35b-a3b-abliterated/index.md): Sparse MoE architecture, upstream Qwen history, and refusal-direction editing.
- [YuYu1015 Ornith 1.0 35B](https://abliterated.cloud/blog/ornith-1-0-35b-abliterated/index.md): Self-scaffolding coding lineage and the corrected weights-only abliteration.
- [Huihui Qwythos 9B Claude Mythos 5](https://abliterated.cloud/blog/qwythos-9b-claude-mythos-5-1m-abliterated/index.md): Qwen architecture, Empero reasoning fine-tuning, long-context caveats, and the Huihui derivative.
- [cebeuq Ornith 1.0 397B W4A16](https://abliterated.cloud/blog/ornith-1-0-397b-abliterated-w4a16/index.md): Streaming refusal-direction surgery and immediate four-bit conversion at 397B scale.

## Exact models and API identifiers

- Qwen3.6 35B A3B — API model ID `huihui-ai/Huihui-Qwen3.6-35B-A3B-abliterated` — $5.45/hour
- Ornith 1.0 35B — API model ID `YuYu1015/YuYu1015-Ornith-1.0-35B-abliterated` — $5.45/hour
- Qwythos 9B Claude Mythos 5 — API model ID `huihui-ai/Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliterated` — $2.34/hour
- Ornith 1.0 397B W4A16 — API model ID `cebeuq/Ornith-1.0-397B-abliterated-W4A16` — $10.90/hour

Compatibility aliases remain available for existing clients: `mn/god`, `mn/code`, `mn/fast`, `mn/ornith-397b`, and `nuri/ornith-397b-abliterated`.

Model weight downloads are free. Managed GPU inference is paid. Automated
billing is not currently live.

## Access

Request private access through Signal:
https://signal.me/#p/+13103408213