Kept only in this page's memory and sent with your request straight through to Anthropic. It's never stored or logged. Each comparison makes 3 small API calls (Haiku, Sonnet and Opus), billed to your key: roughly $0.01 per run.
Laya is an open-source decision model from Convai Innovations. Instead of generating text, it reads the note once and scores every option in a single forward pass. This version was fine-tuned with Snapcall on galileo-ai/medical_transcription_40 (one RTX 4090, 12.5 min). On 496 held-out notes it reaches 64.3% accuracy across 40 overlapping labels, with a calibration error of 0.048. Use its confidence to auto-route clear cases and send uncertain ones to a person.
Laya runs here on a CPU, so latency is higher than on a GPU or a laptop. Claude comparisons use Anthropic API list prices. Not medical advice and not a clinical device. Don't paste real patient data.