decision-machine-1

CloudRaker's hosted decision model, decision-machine-1, sits behind Milliseconds.ai, an API for small, typed decisions on text and images. CloudRaker Inc., based in Montréal, built the API, which returns structured answers such as labels and scores rather than prose.

What decision-machine-1 does

• Yes or no and labels: yes or no answers with a probability, or a label with its full score distribution.
• Ratings and spans: ratings on a scale, and answer spans with character offsets into the source text.
• Structured extraction: fields that match a JSON Schema, named entities, and checks of a value against its source text.
• Compatible endpoint: an OpenAI-compatible interface that supports JSON-schema extraction and tool calls, though not plain chat.

Who it works with

Teams that need typed results from text or images at volume. The vendor advises checking results on the customer's own data and reviewing ambiguous or consequential cases.

Key facts

• Launch: 19 September 2026, the date of CloudRaker's own blog posts.
• Limits: up to 20,000 characters per text and 32 texts per request, with images up to 5 MB in JPEG, PNG or WebP.
• Speed: the vendor quotes about 90 ms of model time for an image decision at medium detail.
• Underlying models: Qwen2.5-1.5B-Instruct handles constrained choices, and GLiNER2.5 handles span extraction.
• Label language: labels and instructions should be written in English.