PRICING

Free to start. Pay per token, or take a plan.

Models on our CPU servers answer free. Priced models answer on GPUs at their price per million input tokens, which an audit sets at no more than $0.04, and output tokens are free. Plans include GPU tokens every month, and a Dedicated GPU keeps one for your organization.

Plans

Free

$0 no card needed

Try every served model, and pay as you go on GPUs.

  • 500 decisions a day on our CPU servers, free
  • $5 of credit once for GPU answers (90 days), claimed on Settings → Billing after you verify your email
  • 1,000,000 requests a minute
  • GPU answers from your credit, at each model's price

Open to everyone

Start free

Team

$49 a month, per organization

For an organization: its owners, admins and developers.

  • 2B GPU input tokens a month for the organization
  • Its owners, admins and developers use them on every model they call
  • Every member uses them on its private models, on GPUs
  • 1,000,000 decisions a month on CPU, 1,200 requests a minute
  • Then pay as you go: its private models from its payer's credit, members' other calls from their own

Granted by our team for now

Contact us

Dedicated GPU

$299 a month, 24 GB

$599 a month for 48 GB

A GPU reserved for your organization, set up by our team.

  • One GPU for your organization's requests, once our team has set it up with you
  • Its private models and the public ones
  • No per-token charges from credit: billed by agreement
  • 3,000 requests a minute

Set up with our team

Contact us

Our team grants plans for now. Your plan, its tokens this month and your credit are on Settings → Billing. Larger volumes or terms of your own: [email protected]. Prices exclude tax. What you buy online is sold by Paddle, our reseller and Merchant of Record; the refund policy says when it is refunded.

Pay as you go on GPUs

$0.04per million input tokens, the most an audit prices a model at. Each model's price is what its GPU costs, measured by an audit on our GPUs, plus our margin.
$0for output tokens. System One models read and decide; they never generate text.
256tokens, the least a request is billed as: per request, not per question. A failed request costs nothing.
Model prices on the inference API
ModelPer 1M input tokensRuns onFree on CPU
supersonic-labs/julia-1$0.01NVIDIA A40Yes
convai-innovations/laya$0.02932NVIDIA A40Yes
convai-innovations/laya-noulxp-gpu$0.04—No
mapika/decider$0.04NVIDIA A40Yes

4 models are priced now, the dearest at $0.04 per million tokens. GPU answers are paid from a plan's included tokens first, then from prepaid credit; the API's usage.billed_tokens says what each request was billed as. How the API works

Credit packs

$10$10.00 of credit, once: at least 250M GPU input tokens.
$25$25.00 of credit, once: at least 625M GPU input tokens.
$50$50.00 of credit, once: at least 1.25B GPU input tokens.
$100$100.00 of credit, once: at least 2.5B GPU input tokens.

In US dollars, paid once: a pack is not a subscription. Its credit never expires. It pays for GPU answers at each model's price per million input tokens, which an audit sets at no more than $0.04. Tax, where it applies, is added at checkout. Packs are sold by Paddle, our reseller and Merchant of Record; within 14 days of buying a pack, the part of its credit you have not used is refunded (see the refund policy). Buy credits on Settings → Billing.

The free tier

Every model our CPU servers hold answers free, within 500 decisions a day. Verify your email, then claim $5 of credit once on Settings → Billing. It lasts 90 days and pays for GPU answers in the API and the playground. Refer a friend and you both get more.

Fine-tuning

Fine-tuning in System One Studio, as it opens, costs what its GPU costs plus 50%, by the started minute. It is paid from bought or granted credit, never from the free credit, and each job holds a cap you choose, so it never costs more.

Questions

How is a GPU request billed?

By its input tokens, at the model's price per million, and at least 256 tokens a request: a request with eight questions is still one request. Output tokens are free, because System One models read and decide without writing text. A request that fails costs nothing.

What does a model cost?

Each model's price is what its GPU costs per million input tokens, measured by an audit on our GPUs, plus our margin; an audit never prices a model above $0.04. Models on our CPU servers answer free within your plan's decisions.

What happens when my plan’s tokens run out?

GPU answers carry on, paid from prepaid credit at each model’s price: yours, or for an organization’s private models its payer’s. Without credit, models on our CPU servers still answer within your decisions. The included tokens renew on the 1st of each month (UTC).

Can I buy a plan today?

Our team grants plans for now; write to [email protected] and say which plan you want. Credit packs ($10, $25, $50, $100) are sold on Settings → Billing, where every account can claim $5 of free credit once its email is verified.

Can I get a refund?

Within 14 days of buying a credit pack, the part of its credit you have not used is refunded (all of it if you have used none; paid credit is spent from your oldest pack first, and a pack bought while paid credit was below zero, after a refund or a chargeback, paid that off first), and a plan bought online is refunded in full within 14 days of its first payment, if none of its included tokens were used. Renewals and credit we give (free, promotional, referral, code or granted) are not refunded. Paddle, our reseller and Merchant of Record, makes refunds to the payment method you paid with. The refund policy has the details.

What is a Dedicated GPU?

A GPU with 24 GB or 48 GB of memory that our team sets up for one organization’s requests: once it is set up, nobody else’s requests queue in front of yours, and until then they run on our shared GPUs. Nothing is charged per token from credit; it is billed monthly, by agreement. Working alone? We set it up on an organization of your own.