Skip to content
Laya Studio
Get an API keyGet a key
Swiss hostingAnswered on GPUs in Switzerland. Nothing you send is stored.
ProductsLaya DecisionsLaya GuardLaya Agent · soon
DevelopersTry it liveQuickstartSwitch from JevPricingStatusAPI reference
LearnWhat is Laya?What is Jev?Fast vs slow AITrustworthy confidenceNo made-up answersAll explainers
CompareLaya vs JevLaya vs ChatGPT-style classifiersJev alternativesSupport triageContent moderationMultilingual intake
ResourcesGlossaryAll use casesBlogFAQ
Pricing
Log in

Pricing

Decision API pricing: 15% below Jev, per input token

$0.0357 per 1M input tokens, against Jev’s $0.042: the same unit, 15% less. 1 credit = 1 input token. Every new workspace starts with 6k free tokens, then you buy credits or pick a monthly plan. Failed requests are never billed.

Start with 6k free tokensEstimate my cost

1 credit = 1 input token

$0.0357

per 1M input tokens, 15% below Jev

To start
6k free tokens
Jev list price, same unit
$0.042 / 1M tokens
Monthly plans from
$19 / mo
Lowest rate (Scale)
$0.03105 / 1M tokens
Swiss-only add-on (+15%)
$0.04106 / 1M tokens
Failed requests
$0
Output tokens
none generated

Same unit as Jev

The same input tokens, 15% less

Laya Studio and TypeSafe Jev both bill the input tokens the model reads, so they compare directly: Jev lists $0.042 per 1M, Laya Studio charges $0.0357. The same request costs 15% less whatever its length.

  • 1 credit = 1 input token. what you buy is exactly what the model reads.
  • No output tokens. Laya generates no text, so nothing else lands on the bill.
  • Plans go lower. down to $0.03105 per 1M on Scale, its volume bonus included.
Compare Laya vs Jev on more than price

Cost calculator

What will AI classification cost you each month?

Move the slider to your traffic. Laya Studio and Jev both bill per input token, so the comparison is like for like: $0.0357 vs $0.042 per 1M input tokens, 15% below Jev. A small LLM used as a classifier is shown as an illustrative example.

1M

One request = one text (a message, an email, a document) you want decisions about.

Average text length
1

e.g. intent, urgency and “is the customer angry?” = 3 questions.

= 265 input tokens per request (each question reads the text once), 265M tokens per month. 1 credit = 1 input token. Every workspace starts with 6k free tokens.

Estimated monthly cost

  • Laya Studio best plan picked$9.46

    $0.0000095 per request · $0.0357 per 1M input tokens · Pay as you go

    265M tokens × $0.0357 / 1M, or the cheaper Starter / Pro / Scale plan ($0.0357 per 1M input tokens)

  • Jev (TypeSafe list price)$11.13

    $0.000011 per request · $0.042 per 1M input tokens

    265M tokens × $0.042 / 1M

  • Small hosted LLM (illustrative)$68.25

    $0.000068 per request · $0.15/M in + $0.6/M out · an assumption, not a quote

    1M requests × (415 in × $0.15 + 10 out × $0.6) / 1M

$1.67 a month less than Jev for the same tokens, and 7× less than the LLM example. Both services bill the input tokens the model reads, so the gap holds at every volume: Laya Studio’s list price is 15% below Jev’s. Your bill depends on how long your texts are and how many questions you ask; a request that fails is free. Choosing on more than price? See Laya vs Jev.
Start with 6k free tokens
Assumptions behind these numbers
  • Laya Studio: $0.0357 per 1M input tokens (1 credit = 1 input token), 15% below Jev’s list price. You are billed the input tokens the model actually reads (reported as usage.input_tokens); like Jev, the text is read once per question, plus ~15 tokens of question wording. Plans include a monthly token allowance with a small volume bonus; tokens beyond it are billed at the list price. The 6k free tokens are a one-time allowance and are left out of the monthly figure.
  • Jev: TypeSafe list price of $0.042 per 1M input tokens, applied to the same token count.
  • LLM: an illustrative assumption, not a quote from any provider: $0.15/M input and $0.6/M output, one call per request with a 150-token prompt, ~15 tokens per question and ~10 output tokens per answer.
  • Token counts per text length are rough averages; real tokenisers vary by language.

Plans

Pick a plan, or just pay as you go

Every option bills input tokens (1 credit = 1 input token). We highlight the one that is cheapest for the 265M tokens a month in your estimate above.

Your estimate: 265M tokens/mo → Pay as you go at $9.46/mo. Change

Best for your volume

Free, then pay as you go

Try it on real traffic.

$0

6k free tokens, once · then $0.0357 per 1M input tokens

6k free tokens, once · 1 dot = 250M

  • All three checkpoints
  • Automatic language routing
  • Playground + usage logs
Start free

Starter

For side projects and first production workloads.

$19/mo

560M input tokens / month · $0.03393 per 1M (+5% tokens)

560M tokens / month · 1 dot = 250M

  • Everything in Free
  • Credits roll over while subscribed
  • Email support
Choose Starter

Pro

For teams running decisions in their core product.

$99/mo

3.05B input tokens / month · $0.03246 per 1M (+10% tokens)

3.05B tokens / month · 1 dot = 250M

  • Everything in Free
  • Credits roll over while subscribed
  • Priority capacity
Choose Pro

Scale

High volume, priority capacity.

$399/mo

12.9B input tokens / month · $0.03105 per 1M (+15% tokens)

12.9B tokens / month · 1 dot = 250M

  • Everything in Free
  • Credits roll over while subscribed
  • Priority capacity
  • Higher rate limits on request
Choose Scale

Need committed-use pricing or a dedicated GPU endpoint? Email hello@laya.studio.

How billing works

You pay for the tokens the model reads

A request carries your text (the state) and your questions. As in Jev, the text is read once per question, so three questions about a 100-token email cost about 3 × (100 + the question’s own wording) tokens. You are billed the usage.input_tokens the model reports.

  • Shorter text, fewer questions. both lower the bill; nothing else is counted.
  • Successful requests only. a request that fails costs nothing.
The request format, in the question types guide Which errors are free, in the error reference

Step through one request

Click through a real-looking request and watch the meter: make the text longer, add the questions, then choose whether it succeeded.

POST /v1/systemone

{
  "state": "Hi, my parcel arrived broken and I want my money back today."(~18 tokens, read once per question)
  "questions": {
"intent": { "type": "choice", "criteria": ["refund", "tracking", "other"] }
"urgency": { "type": "score", "criteria": ["low", "medium", "high"] }
"angry": { "type": "noul", "instructions": "Is the customer angry?" }
  }
}

Step 1 of 5

One request arrives

It carries your text (the “state”) and a list of questions. You pay for the input tokens the model reads: 1 credit = 1 input token. Try making the text longer.

Credits billed (input tokens)

0= $0 at $0.0357 per 1M tokens

1 dot = 8 tokens

Errors are free

Only answered requests are billed

Validation errors (400), out-of-credit responses (402), rate limits (429) and unavailable capacity (503) cost nothing and never use credits.

  • 6k free tokens to start. every new workspace gets 6,000 credits, about twenty short requests; failed requests never use them.
  • Credits keep. purchased pay-as-you-go credits never expire; subscription credits roll over while the subscription is active.
The error reference

Swiss-hosted, every plan

Swiss-only processing: +15%, still below Jev

Every plan, even Free, is answered on GPUs in Switzerland whenever the Swiss pool is up, at the standard rate. Turn on Swiss-only mode (the workspace setting, or the x-laya-residency: ch header) and a request can never leave the country: it costs 15% more credits, $0.04106 per 1M input tokens, still 2% below Jev.

  • GPUs in Switzerland. Our primary inference pool runs on dedicated GPUs located in Switzerland. Every API response tells you where it was processed in the x-laya-region header.
  • Zero content retention. The text and questions you send are processed in memory and discarded when the answer is returned. They are never written to a database or log, and never used to train anything. One exception to know about: the optional chat agent (POST /v1/ask, the MCP ask tool, dashboard Chat) first passes your prompt to a third-party language model to plan the questions, in the US by default and in Switzerland under Swiss-only mode.
  • Your part. You remain responsible for your legal basis to process personal and health data. Laya Studio holds no formal certification (such as ISO 27001) today and does not sign HIPAA BAAs.
How Swiss data residency works

Laya Guard

Guard checks: $0.20 per 1k checks, same balance

Laya Guard judges an agent’s tool call before it runs. It is billed per check, not per token: 5,602 credits each, paid from the same credits, packs and plans as decisions. The same price as mcp-guard.ai.

  • One check, nine scores. a full verdict (allow, ask or block) for one tool call, whatever its length.
  • Batches bill per check. up to 64 checks per request; items that fail are free.
  • Swiss-only +15%. the same add-on as decisions, when the request can never leave Switzerland.
What Laya Guard checks

The agent proposes

db.drop_table({ table: "users" })

The user asked

List the tables in the staging database

Block

P(unsafe) 0.960

  • Destructive0.999
  • Arguments match the request0.214
  • Injected by a tool result0.031
  • Data exfiltration0.021

One check, nine scores, ~15 ms of GPU time.

FAQ

Pricing questions, answered

What am I billed for?
Input tokens: 1 credit = 1 input token, at $0.0357 per 1M input tokens. A successful request is billed the usage.input_tokens the model reports: the tokens it actually encoded. As in Jev, the text (the state) is read once per question, so a request with three questions about a 100-token email costs roughly 3 × (100 + the question’s own wording) tokens. Laya generates no text, so there are no output tokens.
What is free?
Every new workspace starts with 6k free tokens (6,000 credits, 1 credit = 1 input token): about twenty short requests, enough to try it on a few real messages, or one Laya Guard check. Failed requests never use them. After that you add credits or subscribe. There is no other free allowance.
Are failed requests billed?
No. Validation errors (400), out-of-credit responses (402), rate limits (429) and unavailable capacity (503) cost nothing and never use credits. Only successful requests are billed.
Does Swiss-only processing cost more?
Yes, it is an add-on. With Swiss-only processing on (the workspace setting, or the x-laya-residency: ch header on a request), a request costs 15% more credits: $0.04106 per 1M input tokens, still 2% below Jev. Without it, requests are still answered in Switzerland whenever the Swiss pool is up, at the standard rate.
How is Laya Guard billed?
A flat $0.20 per 1k checks: 5,602 credits per check, from the same balance, packs and plans as decisions. A check is one tool call judged on all nine questions, whatever its length. Swiss-only checks cost 15% more; failed checks are free.
Do credits expire?
Purchased pay-as-you-go credits never expire. Subscription credits are added every month (yearly plans too: month by month through the paid year) and roll over while the subscription is active.
How much does an AI classification API cost with Laya Studio?
$0.0357 per 1M input tokens on pay-as-you-go, or a monthly plan from $19 (down to $0.03105 per 1M on Scale, which includes a volume bonus). A one-question decision on a 250-token email is about 265 tokens, so a million of them cost about $9.46.
How does Laya Studio pricing compare with Jev pricing?
Both bill per input token, so they compare directly: Jev lists $0.042 per 1M input tokens (TypeSafe list price) and Laya Studio charges $0.0357, 15% less. The same request costs 15% less on Laya Studio whatever its length. Use the calculator on this page to see both, per request and per month.
Is Laya Studio a cheap alternative to using an LLM for classification?
Usually, yes. An LLM classifier pays for a prompt, your text and the generated answer on every call. Laya answers typed questions in one forward pass without generating text. The calculator shows an illustrative LLM estimate with its formula and assumptions, so you can plug in your own provider’s prices.
Which plan should I choose?
Start with the 6k free tokens. For small or irregular volume, buy a credit pack (from $5). Plans include a monthly token allowance with a volume bonus of up to 15% extra tokens, so they pay off once you use most of the allowance. The calculator highlights the cheapest option for your volume.
What does the chat agent (/v1/ask) cost?
The decision is billed like any other (input tokens, +15% when Swiss-only), plus the planner's LLM tokens: 3 credits per prompt token and 11 per completion token on Mercury 2.5, 18 and 36 on the Swiss planner (Gemma-4 31B) for Swiss-only requests. A failed decision is free.
Is there annual billing, and how do plan changes work?
Yes: every plan is also sold yearly, 12 months for the price of 11 (Starter $209/yr, Pro $1,089/yr, Scale $4,389/yr). The monthly token allowance is added every month of the paid year, not all at once. An upgrade applies immediately: you pay the prorated difference for the rest of the billing period and get the prorated extra tokens. A downgrade applies at the end of the billing period, with no charge and no refund.
Can I self-host instead?
Yes. The Laya weights are Apache-2.0 and pip install laya runs them. Laya Studio is for teams that want a managed, autoscaled endpoint with keys, usage tracking, billing and Swiss processing.
Do you offer volume or dedicated capacity?
Yes. Email hello@laya.studio for committed-use pricing or a dedicated GPU endpoint.

Try it on your own traffic

6k free tokens, then $0.0357 per 1M input tokens — 15% below Jev.

Start freeCompare Laya vs Jev