decision-machine-1
CloudRaker's hosted decision model, served at milliseconds.ai. Two small encoder models read text or one image once and return a typed answer with a probability: yes/no, label, rating, answer span, entities, a filled JSON Schema or a value check. It generates no text.
No weights; the only architecture detail published is "two small encoder models" behind one API plus a scheduler, with no sampling loop. Native routes cover yes-no, classify, classify-tree, rate, answer, extract, entities and verify, each also taking one image (JPEG, PNG or WebP, up to 5 MB). A /v1/systemone route accepts TypeSafe's request shape for noul, choice and score questions; milliseconds.ai states it is not affiliated with TypeSafe AI. Text up to 20,000 characters; batches up to 32 items. $0.04 per million input tokens, output free; test keys get 125M free input tokens a month. CloudRaker reports latency only (0.40 to 0.77 s wall-clock per call including TLS, 36 to 54 ms inference) and no accuracy or calibration figures. The third-party JevBench board (v1.4.2.2) scored it 39.9 (public 0.675, sealed 0.256). SDKs are MIT. Held on 30 September for want of a named maker; CloudRaker's own site, PyPI and GitHub now name it.
What it decides
- noul — answers a yes/no question with one probability
- choice — picks one option from a set
- score — places the input on an ordered scale
- classify — assigns a category from a fixed taxonomy
- extract — pulls spans or fields out of the input
At a glance
| Maker | CloudRaker |
| Released | 2026-09-18 |
| License | proprietary |
Hosted API
Served by CloudRaker (milliseconds.ai) — $0.04/MTok input, $0/MTok output. Get access · API docs.