milliseconds.ai

What it is

milliseconds.ai provides decision-machine-1, a small AI model optimized for bounded tasks like classification, extraction, and yes/no decisions. The API accepts text or images and returns structured data—labels, fields, scores, or answers—that applications can act on directly. It skips token-by-token generation used by large language models, delivering responses in milliseconds for support triage, invoice processing, sales intake, and content moderation workflows.

Questions from the bench

What is decision-machine-1?

decision-machine-1 is a small AI model optimized for bounded tasks like classification, extraction, yes/no decisions, and scoring. It reads text or images and returns structured data that applications can use directly, skipping the token-by-token generation used by large language models.

How is billing calculated?

The API charges $0.04 per million input tokens for production workloads. Input tokens include the validated request body (text, labels, descriptions, hints, questions, types, or schema). Images are billed as a fixed number of tokens based on detail level (low: 1,000, medium: 2,000, high: 4,000). Output tokens are always free.

What is the free tier?

The free tier provides 125 million input tokens per month on test keys with no card required. Test keys support all capabilities but are limited to 30 inference requests per minute and 500,000 input tokens per minute. The free allowance renews monthly but does not apply to production keys.

When should I use decision-machine-1 instead of a large language model?

Use decision-machine-1 for defined classification, extraction, or yes/no tasks where the model can choose from preset labels or extract specific fields. Use an LLM for writing, conversation, multi-step reasoning, or judgments requiring knowledge beyond the supplied text.

What image formats and sizes are supported?

The API accepts JPEG, PNG, and WebP formats with a maximum file size of 5 MB. Images must be sent as bytes, a data URL, or base64. Remote image URLs are not supported, and images are not stored.