meraGPT
Small models for one job; Decider 1 leads typed-decisions
4 followers
Small models for one job; Decider 1 leads typed-decisions
4 followers
Small models that each do one job. Decider 1 answers typed questions about any state as probabilities and is #1 on typed-decisions, Hugging Face's decision benchmark. Restyler 1 makes model output read human. Query Fanout 1 predicts AI searches.
Free Options
Launch Team
Atlassian LoomShow, don't just tell with Loom
Promoted
Since we scheduled this launch, a third model joined: Decider 1. You send it a piece of state (a support conversation, an alert, an invoice, an agent trace) and typed questions about it: yes or no, a choice between labels, or a score on a rubric. Each answer comes back as a probability distribution, and twenty questions over one state are answered in one pass. It is state of the art on typed-decisions, the official Hugging Face benchmark for decision models: 0.768 accuracy and a Brier score of 0.052, against 0.727 and 0.148 for TypeSafe Jev 1.13. It costs $0.04 per million input tokens, output is free, and new accounts get $1 of credit.