31 small models behind one API
Jake Marsh ยท November 19, 2026
Most model calls inside a product are the same handful of jobs. Is this message toxic? What language is it? Is someone trying to hijack the agent? Which queue does this ticket belong in? A frontier model answers those well, slowly, for a few cents each.
Great Work Labs runs 31 small models that each do one of those jobs, behind one API. You send model and input, you get output and the exact cost of the call. A toxicity check is $0.00002. Embeddings are $0.005 per 1,000 texts. Transcripts are $0.001 a minute.
The lineup covers language detection, sentiment, emotions, entities, zero-shot classification, prompt routing, toxicity, content safety, prompt injection, scam detection, grounding checks, embeddings, reranking, JSON extraction, speech to text, OCR, object detection and background removal, among others. Every model page has a playground with real examples and a prompt you can paste into your coding agent.
Billing is prepaid: add credit from $5, it never expires, and you can turn on auto top-up or set a monthly cap. Failed calls aren't charged. Inputs and outputs are never stored.
If you already use OpenAI's embeddings, moderations or transcriptions endpoints, or Cohere's rerank, change the base URL and the key and your code keeps working. Agents can connect to the MCP server without signing up first. Start with the quickstart.