Deep reasoning
Advanced chain-of-thought for complex problem solving, math and logic.
Custom-trained model
Aurorathinks.
Direct, blazing-fast access to aurora — our custom-trained model for deep reasoning, code and natural conversation. OpenAI-compatible and pay per token, with zero data retention.
Uncensored
1M context
Zero retention
OpenAI-compatible
Works with every OpenAI-compatible tool
Capabilities
One model, built for everything you ship.
aurora is custom-trained for advanced reasoning, multilingual understanding, long-context work and agentic tasks — frontier-level performance at a fraction of the cost.
Uncensored
Aurora answers what you actually asked — no arbitrary refusals, no sanitizing, no moralizing. The freedom is the point.
Code generation
Write, debug and explain code across 40+ programming languages.
Long context
Hold entire documents and codebases in a 1M-token context window.
Multilingual
Native-level fluency in English, Chinese, Spanish, Japanese and 20+ more.
Streaming
Tokens arrive as they're generated over Server-Sent Events, so chat UIs feel instant.
The model
auroracustom-trained
One flagship model, trained in-house and served on our own inference stack. Same model and same per-token price for everyone.
Model idaurora
Context window1M tokens
First token<500 ms
Throughput~100 tok/s
Data retentionZero
APIOpenAI-compatible
Request lifecycle
One request in. Tokens stream back. Nothing is kept.
Point any OpenAI client at our base URL. Chat completions and streaming work the way you already expect.
01 authClient key
sk-aurora-••••post
02 gateway
Aurora
meter · route · forgetforward
03 dispatchaurora
gpu-pool / eutokens
04 sse channellive
Platform
Fast where it counts, quiet where it matters.
Optimized inference, transparent pricing and strict privacy — the parts of an AI platform you shouldn't have to think about.
Blazing fast
Optimized inference delivers tokens at speeds rivaling the fastest models on the market, with streaming out of the box.
OpenAI-compatible API
A true drop-in. Change your base URL and key, nothing else.
- POST /v1/chat/completions
- Streaming over Server-Sent Events
- Works with any OpenAI SDK
Private by default
Zero data retention by default. Prompts and completions are never logged in plaintext and never used for training. GDPR compliant.
Built to stay up
A 99.9% uptime target with automatic retries and failover, served from a global edge network for low latency wherever your users are.
Under the hood
What the gateway does on every request.
Metered to the token, auditable by hash, and never stored — shown live from a sample probe.
151ms
Time to first token, sampled liveSpeed
Optimized runtimes behind a controlled gateway for a predictable time to first token.
Input4,096 tkn
Output1,204 tkn
Cost$0.0245
Hash0x4dc1…15be
Prompts not stored
Ledger
See input, output, cached tokens, cost and a one-way request hash — never the raw prompt.
Isolation
Give each app or environment its own key. You can tell traffic apart, and all of them spend one account balance.
FAQ
The short version.
What is Aurora?
A custom-trained large language model, built and served by Aurora API and optimized for reasoning, coding and general-purpose tasks. Call it with the model id aurora.
Is Aurora uncensored?
Yes. Aurora responds without arbitrary refusals or unnecessary moralizing and handles mature themes, sensitive topics and blunt questions directly. You're responsible for using it in accordance with the laws that apply to you and your users.
Is it compatible with OpenAI's API?
For chat, yes: POST /v1/chat/completions follows the OpenAI chat format, including streaming. Point any OpenAI SDK at https://auroramodel.net/v1 with the model id aurora.
Do you train on my data?
Never. Zero data retention by default for API requests. Prompts and completions are not stored after processing.
How fast is it?
Median time-to-first-token under 500 ms, with sustained throughput up to roughly 100 tokens per second.
Can I use Aurora commercially?
Absolutely — every tier allows commercial use, and you own both your inputs and outputs.
How do I pay?
Create a key for free, then top up your account with $25, $50 or $100 in crypto through our Telegram. Credits are added manually after the payment is confirmed in Telegram. Every key you create spends that one balance at the published prices, and all of them stop at $0.
Pricing
One model. One price.
One model. One price.
Pay only for tokens.
Every tier gets the same uncensored model — same prices, same freedom.
Create a key for free and top up your account with crypto in Telegram. Every key you create spends the same account balance until it reaches $0. Credits are added manually after the payment is confirmed in Telegram.
Token prices
Per 1,000,000 tokens — the same for everyoneInput$3.47$0.00347 / 1k
Cached input$0.35$0.00035 / 1k
Output$5.79$0.00579 / 1k
Thinking tokens count as output and are billed at $5.79 per 1M.
Top-up
Top up in Telegram.
Choose $25, $50 or $100 and pay with crypto in our Telegram. The credit is added to your account manually once the payment is confirmed there, and all your keys spend it.
Adds $25.00 to your account≈ 7.8M tokens at an even input / cached / output mix
Pay with crypto
- 1Create a key.
Free and instant. You see the full key once. More keys later share the same balance.
- 2Pay with crypto in Telegram.
We open t.me/AuroraUncensored with your key prefix and amount filled in.
- 3Credit is added manually.
After the payment is confirmed in Telegram, the exact dollars are added to your account.
- At $0 every key stops.
Requests return 402 until credit is added, then all your keys resume.
Plans
Plans for more headroom.
Every account starts on Standard. A plan raises your account's limits and include a one-time usage balance. Token prices are the same on every plan.
Plans are bought in Telegram and activated on your account manually after the payment is confirmed. The plan price is not added to your balance.
API reference
Aurora API documentation
Everything you need to integrate aurora, our custom-trained model. Chat completions follow the OpenAI format — if you've used the OpenAI chat API, you already know how to use Aurora.
Getting started
Aurora is a RESTful API accessible over HTTPS from any environment that can make HTTP requests. Use it via raw HTTP, any OpenAI SDK, or the library of your choice. You'll need an Aurora API key and any HTTP client.
Zero data retention. Prompts and completions are not stored after processing and never used for training. Only token counts and timestamps are kept, for billing.
cURLPythonNode.jsGoLangChainLlamaIndex
Authentication
Every request must include your API key in the Authorization header as a Bearer token. Since Aurora is OpenAI-compatible, pointing any OpenAI SDK at our base URL is all it takes.
Keep your key secret. Never expose it in client-side code or public repositories. If a key is compromised, revoke it from your dashboard immediately.
Available models
One custom-trained flagship model, available to every key.
| Model ID | Context | Description |
|---|---|---|
| aurora | 1,000,000 tokens | Custom-trained flagship — deep reasoning, coding and general conversation. Uncensored: no arbitrary refusals. |
Chat completions
POST /v1/chat/completions — the primary endpoint for interacting with Aurora. It creates a model response for the conversation you pass in messages.
Request body
| Parameter | Type | Req | Description |
|---|---|---|---|
| model | string | yes | Model ID to use. Use aurora. |
| messages | array | yes | The list of messages comprising the conversation. |
| temperature | number | no | Sampling temperature, 0–2. Higher is more creative. Default 1.0. |
| top_p | number | no | Nucleus sampling. Default 1.0. |
| max_tokens | integer | no | Maximum number of tokens to generate. |
| stream | boolean | no | Stream partial deltas as they're generated. Default false. |
| stop | string / array | no | Up to 4 sequences where generation stops. |
| seed | integer | no | For deterministic sampling (best-effort). |
Example request
Response
200 · application/json
Streaming
Set "stream": true to receive tokens as they're generated via Server-Sent Events. The final chunk before [DONE] includes full usage statistics, so you can track consumption even while streaming.
Endpoint
Aurora currently serves one endpoint. Other OpenAI routes are not available yet.
| Method | Path | Purpose |
|---|---|---|
| POST | /v1/chat/completions | Chat responses, with streaming via "stream": true. |
Error handling
All errors return a JSON object with an error field. Retry 429 and 5xx responses with exponential backoff.
| Code | Meaning | What to do |
|---|---|---|
| 400 | Bad request | Check your JSON and parameters. |
| 401 | Invalid or missing key | Verify your Authorization header. |
| 402 | Key balance is $0 | Top up that key — it stops at $0. |
| 403 | Blocked | Check billing, region or account status. |
| 404 | Model not found | Use aurora. |
| 413 | Request too large | Prompt exceeds the context window. |
| 429 | Rate limited | Slow down; retry with backoff. |
| 500 | Server error | Retry with exponential backoff. |
| 503 | Overloaded | Retry with exponential backoff. |
Balance & limits
Your account has one balance, and every key you create spends it. Top-ups ($25, $50 or $100) are paid with crypto in Telegram and added to your account manually after the payment is confirmed. Every request draws the balance down at the published prices, and when it reaches $0 all of your keys return 402 until credit is added. Another customer's balance is completely separate.
Calls stop at $0. Thinking tokens count as output, so set
max_tokens to keep long reasoning in check.Plan limits
| Plan | Keys | Req / min | Tokens / day | / week | / month |
|---|---|---|---|---|---|
| Standard | 1 | 20 | 15M | 105M | 450M |
| Pro | 5 | 60 | 100M | 700M | 3B |
| Plus | 10 | 120 | 500M | 3.5B | 15B |
| Scale | 25 | 200 | 1B | 7B | 30B |
When you go past a limit you get 429. Retry with exponential backoff, or move to a bigger plan.
Best practices
Use prompt caching. Keep your system prompt and static context at the front of
messages. Repeated prefixes are cached automatically and billed at $0.35/1M instead of $3.47/1M — about 90% less.Set max_tokens. Prevent runaway generations and control output cost — the most expensive tier at $5.79/1M. Thinking tokens count as output.
Stream for chat UX. Streaming lowers perceived latency for your users.
Use Aurora responsibly. It responds without arbitrary refusals, so you are responsible for complying with the laws that apply to you and your users.
Aurora
Dashboard
Keys
Your balance, plan and keys, read live from the Aurora database.
Your keys
Full keys are shown once. Right after a key is created, copy it somewhere safe; this dashboard only shows its prefix afterwards. Removing a key from the dashboard doesn't disable it.
Usage
What your keys have spent, from the databaseTop up
Add credit to your account with crypto in Telegram. Every key spends the same account balance.
Choose an amount
Credit is added to your account manually after the payment is confirmed in Telegram.
Redirects to Aurora Support
How it works
- 1Pay in Telegram.
Aurora Support opens with your request ready to send.
- 2We confirm the payment.
Right there in the chat.
- 3Credit lands in your account.
Added by hand. You'll get a notification here.
- At $0 every key stops.
They all resume as soon as credit is added.
Plans
Bought the same way, in Telegram with crypto, and activated on your account by hand after the payment is confirmed. Token prices stay the same.
Your balance and plan update after Aurora Support confirms the payment. This page checks on its own and shows a notification when credit lands or a plan becomes active.
Network
Live connection to auroramodel.net, checked from this browser every 12 seconds.
Checking…Running the first check
Latency now— ms
p50—
p95—
Checks0
Latency mslivep50 —spike
every 12 s
—now
Checks—healthyslowfailed
since you opened this pagenow
Connection
What your SDK should point at- Base URLhttps://auroramodel.net/v1
- Model idaurora
- Key formatsk-aurora-…
- ProtocolHTTPS, OpenAI-compatible
- StreamingServer-Sent Events
What the checks measure
Real round-trips from your browser- Every 12 seconds this page sends a request to auroramodel.net/v1 and times the answer.
- Healthy is under 300 ms, slow is 300 ms to 1 s, and anything slower is a spike. No answer counts as failed.
- History starts when you open the page and isn't stored anywhere.
Guide
From a new key to your first request.
Title
Trust
Your data stays yours.
The highlights of how we handle terms and privacy. You own what you create with Aurora, and we don't keep your prompts.
Terms highlights
- You own your inputs and outputs. We claim no rights over content generated with Aurora.
- Commercial use is allowed with every key.
- Don't use the API for activity that is unlawful in your jurisdiction.
- Don't resell raw API access without permission.
Moderation stance
- Uncensored by default. Aurora doesn't refuse prompts, sanitize answers or add moral commentary.
- You own what you ask and get. Inputs and outputs are yours — and so is the responsibility for using them lawfully in your jurisdiction.
- Mature themes, sensitive topics, blunt answers are all fair game. That freedom is the product.
- The one rule: don't use the API for activity that is unlawful in your jurisdiction.
Privacy highlights
- Zero data retention. Prompts and completions are not stored after processing.
- No training on your data — ever.
- Only usage metadata (token counts, timestamps) is kept, and only for billing.
- GDPR compliant. Data processed in secure data centers.