Questions and answers
Short answers. The docs have the details of the API, and the terms, privacy policy and refund policy are what applies.
The model
What model do you host?
One model: an uncensored build of DeepSeek V4.1 Flash, release 1017. It has two names: deepseek-v4.1-flash-uncensored thinks before it answers, and deepseek-v4.1-flash-uncensored-nothink answers directly, sooner and with fewer output tokens. Both cost the same.
What does “uncensored” mean?
This build has had its refusal training removed, and we add no content filters of our own. It can still decline a request or get things wrong, and its answers can be inaccurate or offensive, so check anything that matters. Our acceptable-use rules still apply: illegal use is prohibited.
How long can a request be?
Up to 1,048,576 tokens, prompt and answer together, and each answer up to 32,768 tokens, reasoning included. With a prompt near the limit, set a smaller max_tokens.
Can it read images or files?
No. The API takes text only: image, audio, document and file content is refused with a 400. There are no embeddings and no image generation.
How is thinking billed?
Reasoning tokens are output tokens, billed at the output price and counted in the usage each answer reports. Use deepseek-v4.1-flash-uncensored-nothink when you don’t need the reasoning step.
Privacy
The privacy policy is the full account.
Do you log my prompts?
No. Prompts and answers are not logged and not used for training; they are processed in memory to answer you. The privacy policy gives the one exception, what the law requires.
What do you keep about each request?
The time, the model, the token counts, the cost and the name of the key used, for billing and your Usage page, with no content. Each record is kept 90 days; daily totals are kept longer.
Is the prompt cache shared with other customers?
No. Each account’s cache is its own, so no other account’s requests can hit it. A cached prompt stays in server memory for a short time and is replaced as new requests arrive.
What do I need to give you to sign up?
A username, an email address and a password. No name, card or phone number: credit is paid in crypto.
Do you use trackers?
No analytics, advertising or third-party trackers. The site sets only the cookies it needs to work.
Pricing and credit
Pricing has a cost estimator and the limits table.
How much does it cost?
$0.18 per million input tokens, $0.01 per million cached input tokens and $0.75 per million output tokens, for both model names. These are launch prices: we email you at least 14 days before any increase.
How does prepaid credit work?
You add credit in US dollars, $5 to $5,000 at a time, and each request is billed from it by the tokens it uses. New requests stop at $0: there is no subscription and no bill afterwards.
Does credit expire?
No, not while your account is open. Credit left on an account you delete is lost.
How does the caching discount work?
It’s automatic. When a request starts with the same tokens as a recent request from your account, such as a system prompt or a long chat, those tokens are billed at the cached price, 94% below the input price, with no charge for writing to the cache.
What are levels?
How many requests your account can run at once grows with the credit it has bought in all: 5 under $20, 15 from $20, 30 from $100, and 60 from $1,000. Your level rises as soon as a top-up is credited (the table).
Can I cap what one key spends?
Yes: give a key a lifetime spend limit when you create it. The Usage page shows each key’s spend.
Coding plans Coming soon
Not on sale yet. This is what they will include; until then, pay-as-you-go is the only way to buy.
What is a coding plan?
A fixed price for 30 days of use, up to a usage cap per 5 hours and per week, for coding agents and other steady work. Usage is counted at the pay-as-you-go prices, over a rolling 5 hours and a rolling 7 days, so a large prompt counts for what it is.
| Plan | Per 30 days | Usage per 5 hours | Usage per week | At once |
|---|---|---|---|---|
| Starter | $20 | $1.50about 500 agent requests | $7.50about 2,400 agent requests | 3 |
| Plus | $50 | $3.60about 1,200 agent requests | $18about 5,900 agent requests | 6 |
| Pro | $100 | $7.20about 2,400 agent requests | $36about 11,800 agent requests | 12 |
| Max | $200 | $14.40about 4,700 agent requests | $72about 23,500 agent requests | 24 |
What is an “agent request”?
Our unit for the table: a coding-agent turn with a 125,000-token prompt, 95% of it cached, and 1,000 tokens out, which costs about $0.003 at the prices above. Yours may be larger or smaller; the cap is in dollars of usage, not in requests.
What happens when I reach a cap?
Plan requests get 429 with code plan_limit_reached and a Retry-After for when enough usage has aged out of the window. Plan usage never touches your prepaid credit, and your pay-as-you-go keys keep working if you have credit. A plan that has ended answers 402 with code plan_ended until it is renewed.
Are plan requests served first?
When the servers are busy, plan requests go ahead of pay-as-you-go requests, and higher plans ahead of lower ones. We don’t promise uptime figures.
How do I use a plan?
A plan has its own API keys, separate from your pay-as-you-go keys and labelled on the Keys page. They work like any other key and count toward the same 10-key limit.
How do I pay for and renew a plan?
Pay a crypto invoice from the Billing page, or pay from your prepaid credit if it covers the price. Crypto can’t be charged automatically, so a plan renews from credit if you leave that on and your credit covers it; otherwise we email a payment link 3 days and 1 day before it ends. An unpaid plan lapses: its keys stop working until you pay again, and your data is kept.
Can I change or cancel a plan?
An upgrade starts at once, and the old plan’s unused days go back to your prepaid credit; a downgrade starts at the next renewal. Cancelling stops renewal, and the plan runs to the end of what you paid. Refunds follow the refund policy.
Payments
Every case is in the docs.
How do I pay?
In Bitcoin, Litecoin, Monero or USDT, by paying an invoice for the US dollar amount you choose. There’s no card on file and nothing recurring.
How long is an invoice valid?
It fixes the exchange rate for 15 minutes. You can have 3 unpaid invoices open at once.
When does the credit arrive?
Once the payment confirms on its network. A payment sent in time that confirms after the invoice has expired is still credited when it confirms.
What if I pay too little?
A shortfall of a cent or less counts as paid in full. Otherwise send the rest to the same invoice while it is open; if it expires part paid, what you sent in time is credited at its exchange rate, rounded down to the cent.
What if I pay too much?
The invoiced amount is credited. The extra is checked by hand and refunded on request, as the refund policy describes.
What if I pay after the invoice expired?
It isn’t credited automatically: we check it by hand, then credit it or refund it. Don’t pay an expired invoice; start a new top-up.
Can I get a refund?
In some cases. The refund policy says when, how to ask and how refunds are paid.
My payment isn’t on my balance
If it isn’t there within a day, contact support with the invoice ID, shown under Payments on the Billing page.
Using the API
The docs have examples, every limit and every error.
Which SDKs work?
The OpenAI SDKs, with base URL https://api.lunomi.io/v1, and the Anthropic SDKs, with base URL https://api.lunomi.io. One key works with both, and streaming and tool calls work as usual.
Can I use it with Claude Code, Cursor or other tools?
Any tool that lets you set a base URL and an API key. Coding agents on the Anthropic API, Claude Code among them, read ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN; Cursor, Continue, Open WebUI and similar apps take a custom OpenAI-compatible provider. Apps and tools has the settings.
Which endpoints are there?
POST /v1/chat/completions (OpenAI), POST /v1/messages (Anthropic) and GET /v1/models. There are no embeddings, Responses API, batches, files, images or audio.
What are the rate limits?
Your account can run a number of requests at once, set by its level, and a number a minute; past either, a request gets 429 with a Retry-After header. The OpenAI and Anthropic SDKs retry twice by themselves; raise max_retries to wait longer (Limits).
What do the errors mean?
401 is a wrong or revoked key, 402 means your credit or the key’s spend limit can’t pay for the request, 429 is a limit, and 503 asks you to retry after Retry-After. Errors lists every message and what to do.
Account and security
How does sign-in work?
With your password and a code we email you, each time; the code expires in 10 minutes and works once. You stay signed in for up to 7 days, or 12 hours without use, and every sign-in and change is in your account’s security activity.
Can I sign in with Google?
Not at the moment: sign-in uses a password and an emailed code.
How do API keys work?
A key is shown once, when you create it, and up to 10 can be active at once, all drawing on the same balance. Use one per app, so you can revoke one alone; a revoked key is refused from its next request.
I forgot my password
Reset it from the sign-in page: we email a code to the address on your account.
How do I delete my account?
From the Account page, at any time. Deleting it ends your access, and any credit left on it is lost. What we keep afterwards, and for how long, is in the privacy policy.
Another question?
Ask on the Support page, or email [email protected].