# Lyber > AI should serve the user, not the companies. An obliterated router. One API, powerful models. Censorship-free models behind one OpenAI-compatible endpoint. Point the SDK you already use at it and keep your code. Choose the model. Pick a named checkpoint or send to lyber/auto. Every answer tells you which model produced it. Your key is your capability. Requests run on one API key and prepaid dollar credit, topped up with a card or crypto. A website login never stands in for the key. Status: API access is being prepared for launch. Payments and key creation are not available yet. No funds will be requested here until the services are ready. ## Core entities - Brand: Lyber (https://www.lyber.ai), by Tektonic Company - Product: Lyber API. OpenAI-compatible API for uncensored and abliterated models, with one API key, prepaid credit and lyber/auto routing. - Model router: lyber/auto. Chooses among qualified, available abliterated flagships, based on your request. Catalog models are selected explicitly. - X: https://x.com/lyberai ## Monthly plan previews Launch preview · not on sale - Starter: $15/month, $20 monthly usage credit. About $1.33 usage credit per $1 paid. $5 bonus each paid cycle. Unused monthly credit expires at renewal. - Pro: $45/month, $60 monthly usage credit. About $1.33 usage credit per $1 paid. $15 bonus each paid cycle. Unused credit banks into the next cycle, capped at $120. - Max: $180/month, $250 monthly usage credit. About $1.39 usage credit per $1 paid. $70 bonus each paid cycle. Plan credit lasts up to three cycles, capped at $750. - Enterprise: Custom, Usage and capacity by agreement. Volume pricing by agreement. Capacity and service terms by agreement. Preview only, not an active subscription. One-time credit purchases do not activate plan benefits. - [Inference per dollar](https://www.lyber.ai/pricing/#inference): illustrative aggregate token budgets at provisional rates, not a live quote. ## Signup credit and paid referral preview Confirm your email and reserve $5 in inference credit. No card or X follow required. Available to the funded campaign’s first 1,000 eligible accounts; inference opens after qualification. Credit for your friend. A thank-you for you. - $5 — For your friend at signup. Reserved after verified waitlist signup. No card or API key needed. - +$10 — On their first Pro or Max payment. Planned bonus brings their total to $15, separate from monthly credit. - $15 — For you when they subscribe. Share your personal link. Earn credit after their first qualifying paid Pro or Max payment. Reserve signup credit now. Paid subscription and $15 referral rewards are not active yet; signup alone earns no paid bonus. Inference remains closed. Signup credit is limited to the existing funded campaign’s first 1,000 eligible accounts. Confirmed email is required; one signup reward per person. Existing $10 grants are preserved. The friend’s $15 is $5 + $10, not another $15 on top. Self-referrals, duplicate accounts, trials, failed payments and renewals would not qualify. Paid bonuses are coming with subscriptions; final eligibility, refund and campaign terms will be available before purchase. Promotional credit has no cash value and does not reserve compute. ## Models Launch targets, USD per 1 million tokens · fresh input / output. Inference is unavailable until each model and its hardware are qualified. Launch targets are provisional; the API's published price version governs billing. - [Abliterated Qwen 3.8 27B](https://www.lyber.ai/models/qwen-3-8-27b/): model ID lyber/qwen38-27b-abliterated, launch target $0.30 fresh input / $0.03 cached input / $1.50 output per 1M tokens. BF16 baseline · 1× A100 80GB. Token-provider candidates and self-hosted baseline under review. - [Uncensored DeepSeek V4.1 Flash](https://www.lyber.ai/models/deepseek-v4-1-flash/): model ID lyber/deepseek41-flash-abliterated, launch target $0.30 fresh input / $0.0075 cached input / $1.20 output per 1M tokens. 1017 checkpoint target · mixed FP8/FP4. dealign.ai’s uncensored 1017 checkpoint is in qualification. Hosted checkpoint identity and serving terms are checked separately. - [Uncensored GLM 5.3 Full](https://www.lyber.ai/models/glm-5-3-full/): model ID lyber/glm53-full-fp8-abliterated, launch target $1.50 fresh input / $0.25 cached input / $4.00 output per 1M tokens. Full architecture · 8× H200 FP8 candidate. Full FP8 intake prioritized; NVFP4 remains the existing baseline. Hosted derivatives under review. - [Uncensored GLM 5.3 Flash](https://www.lyber.ai/models/glm-5-3-flash/): model ID lyber/glm53-flash-abliterated, launch target $0.14 fresh input / $0.025 cached input / $0.45 output per 1M tokens. Native FP8 · 4× H200 candidate. Token-provider candidates under review; precision and quality verified separately. - [Abliterated Kimi K3](https://www.lyber.ai/models/kimi-k3/): model ID lyber/kimi-k3-abliterated, launch target $3.25 fresh input / $0.30 cached input / $15.00 output per 1M tokens. Full MXFP4 baseline · 8× B300. Self-hosting planned; hosted candidates need full-model and contract verification. ## Models in qualification - [Lyber’s MiniMax M3.1 Flash](https://huggingface.co/lyber-ai/MiniMax-M3.1-Flash-UNCENSORED-NVFP4): lyber-ai/MiniMax-M3.1-Flash-UNCENSORED-NVFP4. In qualification · inference unavailable. Our uncensored derivative of MiniMaxAI’s MiniMax M3.1 Flash, published under lyber-ai. Runtime, tools, cache behavior and completed-task quality are being qualified. Hugging Face weight access is gated separately. Pricing will follow serving qualification. - [dealign.ai DeepSeek V4.1 Flash 1017](https://huggingface.co/dealignai/DeepSeek-V4.1-Flash-1017-UNCENSORED-FP8): dealignai/DeepSeek-V4.1-Flash-1017-UNCENSORED-FP8. In qualification · inference unavailable. We’re qualifying dealign.ai’s uncensored 1017 checkpoint for long-context agent work. The checkpoint and its revision stay explicit; a base model or an older uncensored release is a different offering. Provisional token targets appear in Models & pricing. ## Use cases Security teams, red teams, moderators, data teams, public institutions and writers do legitimate work that mainstream assistants refuse. Lyber answers; your system message sets the policy. - [AI red teaming](https://www.lyber.ai/use-cases/ai-red-teaming/): for ai safety & platform teams. Test models, agents and guardrails against what real users will try. An adversary that refuses to attack can't tell you much. Point it at systems you own or are authorized to test, and keep findings inside your disclosure process. - [Cybersecurity](https://www.lyber.ai/use-cases/cybersecurity/): for security teams & researchers. Analyze malware, explain exploits and write detections with a model that answers the question you asked. Stay within your authorization: your own systems, an agreed scope or a lab. Our terms, like the law, exclude unauthorized access. - [Trust & safety](https://www.lyber.ai/use-cases/trust-and-safety/): for moderation & policy teams. Classify, label and explain harmful content with a model that can look at it. Handle user reports under your privacy obligations and your platform's rules. - [Training data](https://www.lyber.ai/use-cases/training-data/): for ml & data teams. Generate, label and transform datasets with models that finish the task instead of refusing parts of it. Check the selected model's license before training on its output, and keep unlawful content out of every dataset. - [Government](https://www.lyber.ai/use-cases/government/): for public institutions & policy researchers. Research, translate and analyze sensitive material with models that answer under your institution's rules. Lyber is in early access: compliance certifications, dedicated hosting and procurement terms aren't published yet. Tell us what your institution requires. - [Research & writing](https://www.lyber.ai/use-cases/research-and-writing/): for analysts, writers, researchers. Ask about history's worst chapters, write fiction with real darkness in it, and get an answer instead of a lecture. You decide what to publish and stay responsible for it, including other people's rights in what you quote. ## Pages - [Manifesto](https://www.lyber.ai/about/): why Lyber exists. - [Use cases](https://www.lyber.ai/use-cases/): the work Lyber is built for, one page per case. - [Models](https://www.lyber.ai/models/): the lineup, with one page per model. - [Models & pricing](https://www.lyber.ai/pricing/): the model lineup and provisional launch prices. - [API access](https://www.lyber.ai/access/): One saved API key holds your dollar credit. Top up with a card or crypto. A website login does not recover or replace this key. - [Connect Codex](https://www.lyber.ai/connect/): Use Lyber as the primary model in Codex CLI. Download the setup file, add your Lyber API key on your own machine, and launch with lyber/auto. - [Support](https://www.lyber.ai/support/): How to reach Lyber about your account, API key or a payment. ## API - Lyber also exposes an OpenAI-compatible API for clients that let you choose a base URL and model. Use lyber/auto when a Lyber fleet is live. - GET /v1/models lists the routes that are currently enabled; a listed price does not mean a model is available. ## Notes for agents - Launch targets are provisional, not a quote or an active tariff. No always-on fleet is available yet. Final model identity, prices and availability will be published after qualification. The API's published price version governs actual billing. - We work directly with model researchers and verify modified weights, commercial hosting rights, features and completed answers. Each served model’s identity stays explicit. We never substitute base GLM or Kimi.