End-to-end confidential · TEE-attested

Private
intelligence you own.

Your prompts leave nothing behind: not stored, not trained on, verifiable in hardware. GLM-5.3and the leading open models, OpenAI-compatible. Developers self-serve. Companies get privacy they can audit.

ConfidentialityTEE · SEALED0 bytes retained
Deploy GLM-5.3Try the private chatFor companies →
Urgent · the state of AINetwork view · liveClosed // centralizedAleph Cloud // distributed[ figures illustrative ]
⚠AI today · closed and centralized

Their models.
Their rules.

A handful of closed labs run the AI you depend on. Prompts are stored, profiled and trained on. Access can be re-priced or revoked overnight, and anything they keep can be subpoenaed.

Open models · nothing retained

Your data is yours.

Inference on a distributed network that never holds your data. Open models, lower prices, no lock-in.

Hover the nodes
Prompts stored
4,271,882,090,113 bytes
Confidentiality
None · logged & profiled
Flagship model · price
$27.40 / 1M · unstable
OpenAI-compatible✦No KYC✦TEE-attested✦Nothing stored✦Subscribe or pay per token✦Company agreements✦Powered by Aleph Cloud✦
OpenAI-compatible✦No KYC✦TEE-attested✦Nothing stored✦Subscribe or pay per token✦Company agreements✦Powered by Aleph Cloud✦
( 01 ) · The models

One endpoint.
The leading open models.

GLM-5.3 leads; the rest of the catalog is one parameter away, on the same key.

GLM-5.3Flagship
ToolsReasoning
Online
$1.40
Input · 1M tokens
$4.40
Output · 1M tokens
256K
Context window
~120
Tokens / second

SubscribeFixed price

A monthly plan with a predictable bill. Same endpoint, same models.

CreditsPay per token

No commitment, no minimum. Usage past a subscription bills at the same per-token rate.

Deploy GLM-5.3 in a minute.

Free to start. No card, no KYC. Keep your OpenAI SDK, change one line.

Get your API key Read the docs
8 more open models on the same endpointOutput $ / 1M+
Qwen3.6-35B-A3B$0.15 / $0.50
Qwen3.8-27B$0.15 / $0.50
Qwen3.8-27B (TEE)$0.15 / $0.50
GLM-5.3-Flash$0.15 / $0.50
Hermes 3 8B (TEE)$0.15 / $0.60
Qwen3.5-122B-A10B$0.25 / $1.75
DeepSeek V4 Flash$0.25 / $1.75
DeepSeek V4.1 Flash$0.25 / $1.75
( 02 ) · Plans & pricing

Start free. Scale when you do.

Subscribe for steady usage, or pay per token with credits. Card or crypto.

Free
To try it out
€0
Open chat →
Included
  • The private chat, no account
  • Open models
  • No card required
Go
For everyday use
€8/mo
Subscribe →
Everything in Free, plus
  • Access to all models
  • Much higher usage than Free
  • Longer sessions before a cooldown
Plus
For heavy, daily workloads
€20/mo
Subscribe →
Everything in Go, plus
  • 3× higher weekly limit than Go
  • Comfortable for all-day work
  • Long documents and coding sessions
Max
For power users and automation
€100/mo
Subscribe →
Everything in Plus, plus
  • 10× higher weekly limit than Plus
  • Built for intensive, sustained use
  • Headroom for long-running agents

Prices incl. VAT. Beyond your plan, pay per token with prepaid credits. Chat is free to start; LiberClaw agents have their own free tier.

( 03 ) · Cost vs. benchmarkJul 2026

A fraction of the price.

Open models on distributed infrastructure, without a hyperscaler margin. Comparable benchmarks, up to 85% less per token.

LibertAIGLM-5.3$4.40Bench 91
AnthropicOpus 4.8$25.00Bench 93
OpenAIGPT-5.5$30.00Bench 94
$ / 1M output tokens · bench: GPQA Diamond, % correct
( 04 ) · Trust model

Private by proof.

Most providers ask you to trust a policy. LibertAI is built so there's nothing to trust: the infrastructure can't remember you. The same guarantees hold for a side project and a regulated company.

CommitmentsWhat we do

Built by developers, on open models and open standards.

✓Support developers+
OpenAI-compatible APIs, honest docs and open models you can inspect. Keep your SDK, prompts and stack.
✓Cut inference costs+
Lower per-token prices than the closed flagships, and up to $10,000 in credits for early-stage teams.
✓Keep you in control+
No lock-in, no gatekeeping, no KYC. Switching out takes the same one line as switching in, and everything runs on a decentralized network you can verify.
GuaranteesWhat we don't do

Privacy comes from the architecture, not a promise.

✕No retention+
Requests are processed and forgotten. No prompts, responses or IP addresses in our database, and nothing to subpoena.
✕No training on your data+
Your prompts never become weights, and are never used to train, tune or evaluate any model.
✕No blind trust+
Sensitive workloads run in Trusted Execution Environments with remote attestation. Check the enclave yourself, or hand the attestation to your security team.
( 05 ) · Migration

Start, or switch,
in a minute.

Fresh project or existing stack, your OpenAI code already works. It just talks to a different address.

01

Create an account

Email or wallet, no KYC.

02

Generate an API key

One click in the console.

03

Pick GLM-5.3

Or any model in the catalog.

04

Point your client here

Keep your SDK and prompts.

THE ENTIRE MIGRATIONdiff
- base_url="https://api.openai.com/v1"
+ base_url="https://api.libertai.io/v1"
( 06 ) · The sovereign stack
Two particles become linked, so measuring one instantly determines the other.Why is the sky blue?Shorter blue wavelengths scatter more off air molecules than red.
ChatExplore →
Audit auth middleware[running]
Rotate session tokens[running]
Find stale references[done]
Thought for 4.1s · persistent memory
AgentsExplore →
$ libertai code
→ launching claude-code
→ backend api.libertai.io · glm-5.3
→ private session · 0 logs
▌
CodeExplore →
search the live web, privately
From $0.002 / query · pays via x402
SearchExplore →
For companies

Private AI for your company.

Pilots, custom agreements, volume terms, EU invoicing — the same confidential stack, with our team behind it.

Contact us to get our help →
( 07 ) · Startup program

$10,000 to build private.

For startups shipping on GLM-5.3 and the full catalog, with migration help from our team. Larger company?See custom agreements →

Apply for creditsReply within 48h
AmountUp to $10,000
EligibilityEarly-stage teams
UseInference credits
ReviewCase-by-case
( 08 ) · FAQ
How do I know it is actually private?+
Every model runs with nothing stored and nothing trained on your data. For sensitive workloads you do not have to take our word for it: TEE-served models like Hermes 3 8B run inside hardware-sealed enclaves on Aleph Cloud that even the node operator cannot inspect, with cryptographic attestation you can verify yourself.
Will my existing code work?+
Yes. The API is OpenAI-compatible: keep your SDK, swap the base URL and key. The docs cover streaming, tool use and vision.
How does billing work?+
Two ways: a subscription for regular usage, or prepaid credits billed per million tokens. Usage past a plan's allowance bills at the same per-token rate. Card or crypto, no seats, no KYC.
Can we use LibertAI as a company?+
Yes. Same API, same guarantees, plus custom agreements on request: volume terms, higher rate limits, direct support and EU invoicing. See the Services page.
Is there a free way to try it?+
Yes. The chat at chat.libertai.io is free, no account needed. LiberClaw's free tier runs two always-on agents, no card required. The API has no minimum: top up a few dollars and pay only for the tokens you use.
What happens if a node goes down?+
Workloads reschedule across the network. Your endpoint is not one company's datacenter: it is a distributed pool of Aleph Cloud nodes, so no single machine, provider or jurisdiction can take it offline.
Which models can I run?+
GLM-5.3 leads, with open models behind the same endpoint: Qwen, Gemma, DeepSeek, Hermes and more, plus image generation, embeddings, speech and private web search. All open-weight, so nothing can be deprecated out from under you.

Own your
intelligence.

Deploy GLM-5.3Read the docs
Nothing stored. Nothing trained on. One line to switch.