Built for the work other LLMs refuse.
Abliterated frontier models, served through the OpenAI and Anthropic APIs.
Read the docsModels without filters
Abliterated models are served without a separate content filter.
No conversation storage
Your prompts and the responses they generate are not stored.
Bring your own tools
Use
Claude Code,
Codex, OpenCode, or any OpenAI- or Anthropic-compatible client.Built for sustained work
Run extended coding sessions and support authorized security research and red-team exercises.
Read as it writes
Deliver responses chunk by chunk over server-sent events.
Connect your functions
Models return structured calls to functions you define.
Reuse prompt prefixes
Matching prefixes are cached automatically and billed at the cached input rate.
Match the model to the job
Choose Kimi, DeepSeek, or GLM to suit your workload and budget.
Abliterated models in practice
Use these models for security research, authorized red-team exercises, and testing systems you control. Your inputs and model outputs remain yours.
View usage termsThreat analysis and detection
Analyze suspicious software, reverse-engineer binaries, and build detections from what you find. For this work, start with Kimi K3.1 Pro.
Explore security workflowsRed-team assessments
Use models in authorized penetration tests and red-team exercises to map gaps in defenses, build test tools, and document findings.
Read the assessment guideSynthetic datasets
Generate training data and evaluation sets in bulk, including edge cases other models may skip, while keeping per-token costs low.
Explore dataset generationAgent-assisted coding
Keep coding sessions moving in
Configure Claude Code
Claude Code,
Codex, or OpenCode with fewer interruptions from model refusals.
Connect to 
Use your existing SDK, coding agent, and prompts. Update the API base URL and choose a model.
Abliterated models with 1M tokens of context
Zero data retention
Your prompts and model responses pass through our servers. We discard both after processing.
Trust centerFrequently asked questions
Contact usModels and setup
What is an abliterated model?
Abliteration changes a model’s weights to reduce built-in refusals. Ablivra serves these models as modified, without additional content filters or moderation.
Who uses Ablivra?
Security researchers, red teams, trust and safety teams, AI teams, and developers who work with coding agents use Ablivra for tasks that general-purpose models may decline.
Can I use the tools I already have?
Yes. Ablivra supports OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages. You can use
Claude Code,
Codex, OpenCode, DeepSeek Harness, Kilo Code, Cline, the official OpenAI and Anthropic SDKs, or a client that accepts a custom OpenAI base URL.
How do I connect my existing SDK?
Keep your OpenAI or Anthropic SDK. Use the base URL in the quickstart, add your Ablivra API key, and select the model you want to run.
Read the quickstartHow much context can I send?
Each model accepts up to 1,048,576 tokens of context. Check the model documentation for its output limit.
Accounts and billing
Can my company share an account?
Yes. Invite your team to a shared account and let each member create their own API keys. Security research, authorized red-team exercises, and testing your own systems are allowed under our terms. Your inputs and model outputs belong to you.
Can I set spending limits for my team?
Teammates share a balance or plan. You can set a daily limit, an overall limit, and an expiry date for each API key. View spending by person, key, or model, export a CSV, and use the activity log to see who made changes.
What data do you retain?
We discard prompts and model responses after processing. We retain billing records, including token counts and request costs.
How do credits and monthly plans work?
Prepaid credits start at $5 with no monthly minimum. Monthly plans start at $10. We bill each request by the token at the published model rates.
How can I pay?
We accept cards through Stripe, plus Monero, Bitcoin, Ether, USDT, and USDC.
Who can I contact about higher volumes?
Send us a short description of your workload and expected volume through the contact page. Our team reads every message and replies within one to two business days.



