Small models for specific jobs. A fraction of a cent per call.
Moderation, PII redaction, embeddings, transcripts, OCR, JSON extraction and 21 more task models behind one API. Each one does one job well, answers in milliseconds, and costs from $0.000005 a call.
$ curl labs.greatwork.company/v1/run -H "Authorization: Bearer $GWL_API_KEY" -d '{"model":"veil-1","input":"Hi, I'm Dana Whitfield, my account email is dana.whitfield@example.com and my card ending 4242 4242 4242 4242 was charged twice. Call me at (512) 555-0187."}'
{
"redacted": "Hi, I'm [PERSON], my account email is [EMAIL] and my card ending [CREDIT_CARD] was charged twice. Call me at [PHONE].",
"spans": [
{
"type": "person",
"text": "Dana Whitfield",
"start": 8,
"end": 22,
"score": 0.9794
},
{
"type": "email",
"text": "dana.whitfield@example.com",
"start": 44,
"end": 70,
"score": 0.9949
},
{
"type": "credit_card",
"text": "4242 4242 4242 4242",
"start": 90,
"end": 109,
"score": 0.9988
},
{
"type": "phone",
"text": "(512) 555-0187",
"start": 140,
"end": 154,
"score": 0.9987
}
],
"counts": {
"person": 1,
"email": 1,
"credit_card": 1,
"phone": 1
}
}$ curl labs.greatwork.company/v1/run -H "Authorization: Bearer $GWL_API_KEY" -d '{"model":"sieve-1","input":"You're a complete idiot and everyone here knows it."}'
{
"flagged": true,
"labels": [
"toxicity",
"insult"
],
"scores": {
"toxicity": 0.9095,
"severe_toxicity": 0.0412,
"obscene": 0.1594,
"threat": 0.0045,
"insult": 0.9069,
"identity_attack": 0.0133,
"sexual_explicit": 0.0032
}
}$ curl labs.greatwork.company/v1/run -H "Authorization: Bearer $GWL_API_KEY" -d '{"model":"mold-1","input":"Hi team, this is Maria Gonzalez from Acme Logistics. Our invoice INV-2291 for $4,380.50 is due on March 3rd 2026. Call me at (415) 555-0137 if anything looks off.","schema":{"type":"object","properties":{"name":{"type":"string"},"company":{"type":"string"},"invoice_number":{"type":"string"},"amount":{"type":"number"},"due_date":{"type":"string","description":"ISO date YYYY-MM-DD"},"phone":{"type":"string"}}}}'
{
"data": {
"name": "Maria Gonzalez",
"company": "Acme Logistics",
"invoice_number": "INV-2291",
"amount": 4380.5,
"due_date": "2026-03-03",
"phone": "(415) 555-0137"
}
}$ curl labs.greatwork.company/v1/run -H "Authorization: Bearer $GWL_API_KEY" -d '{"model":"atlas-1","input":["How do I reset my password?","Où puis-je télécharger ma facture ?","Shipping takes 3 to 5 business days."]}'
[
{
"embedding": [
0.074629,
0.000765,
-0.006023,
-0.084802,
0.053536,
-0.06474,
"... 378 more"
],
"dims": 384
},
{
"embedding": [
0.038047,
-0.003843,
-0.047707,
-0.092392,
0.030254,
-0.007609,
"... 378 more"
],
"dims": 384
},
{
"embedding": [
-0.005012,
-0.005834,
-0.02256,
-0.060305,
0.045396,
-0.036372,
"... 378 more"
],
"dims": 384
}
]Pick the job, not the model size.
Every model has the same request shape, the same error format and a playground with real examples. Name the model, send input, get output.
Understand text
Language, sentiment, emotions, entities, topics and your own labels.Safety and trust
Toxicity, policy violations, prompt injection, PII, scams and unsupported claims.Search and retrieval
Embeddings, reranking, keyphrases and emoji for search, RAG and recommendations.Write and restructure
Punctuation, summaries, titles and typed JSON from messy text.Audio
Speech to text with word timestamps.Vision
Tags, text, objects, cutouts and NSFW checks for images.Most AI calls in a product are the same ten jobs.
Is this toxic? What language is it? Strip the emails out. Which ticket queue? A frontier model answers those well, slowly, for a few cents each. A model trained for that one job answers them in milliseconds for a thousandth of that.
Priced per call
Classifiers cost $0.02 per 1,000 texts. Every response says what it cost, and you're never billed more than the estimate.
One request shape
model + input + options, always. Batches are arrays. Outputs share conventions: label and score, spans with offsets, boxes in pixels.
Nothing kept
Inputs and outputs are never stored or used for training. We keep the metadata a bill needs: model, units, cost, time.
Fast enough to inline
Classifiers answer in tens of milliseconds, so you can run them on every message, upload or tool result instead of sampling.
Drop-in endpoints
OpenAI-style /v1/embeddings, /v1/moderations and /v1/audio/transcriptions, and Cohere-style /v1/rerank. Change the base URL and key.
Prepaid, with a ceiling
Top up from $5. Turn on auto top-up if you want it, set a monthly cap if you don't. Failed calls aren't charged.
Your coding agent can wire these in for you.
Every page has a markdown twin, there's an llms.txt and an OpenAPI spec, and each model page has a prompt you can paste into Claude Code, Cursor or Codex. Agents can also call the models themselves over MCP: connect once, and the first paid call sends a top-up link.
# Claude Code
claude mcp add --transport http labs https://labs.greatwork.company/mcp
# Or paste into any coding agent:
Use Great Work Labs to redact PII from support
tickets before we send them to the LLM. Docs:
https://labs.greatwork.company/models/veil-1.mdQuestions
How cheap is a fraction of a cent?
A toxicity check is $0.00002. A thousand of them is two cents. Summaries and JSON extraction cost more ($0.0003 to $0.0004 each) because they generate text. Every price is on the pricing page, and every response includes its exact cost.
Do you store what I send?
No. Inputs and outputs live in memory for the length of the call and are never written to disk, logged or used for training. We keep the model name, unit count, cost, latency and status so we can bill you and show your usage.
Is there a free tier?
No. There's no subscription either: you add credit (from $5), it never expires, and calls draw it down. You can browse the playground's examples without an account; they're real outputs we captured on our servers.
What happens when my balance runs out?
Calls return HTTP 402 with a link to add credit. If you turn on auto top-up, we charge your saved card before that happens, and also mid-call if a request would otherwise be refused.
Can I use these from an agent?
Yes, two ways. Agents can write code against the API (every page has a markdown version and there's an llms-full.txt), or call the models directly over MCP at https://labs.greatwork.company/mcp.
Where do the models come from?
Each model page names the base model it's built on and its license. Some we fine-tuned or wrapped with our own logic (chunking long texts, tuned thresholds, language detection, JSON-schema constrained output), and all of them run on our servers.
What about rate limits?
25 requests a second per account to start, and most text models take batches of up to 64 inputs per request, so that's 1,600 texts a second. Write to us if you need more.