Glass for Developers | Clinical Intelligence API

Open original

Glass is offering its clinical AI as a developer API, so other products can embed cited medical answers, differentials, and scribing into their own interfaces. The page is mostly a sales pitch, but it lays out the endpoints, sign-up steps, and pricing in concrete terms.

  • One Messages endpoint returns evidence-grounded clinical answers with inline citations, and a Scribing endpoint turns encounter audio into a diarized transcript and structured SOAP-style note in one call.
  • Pricing starts at $250 per month with usage included, then $3/$16 per million input/output tokens for Glass 5.5 (or $10/$50 for 5.0), plus $0.85 per hour of transcribed audio.
  • Rate limits are 100 requests, 200,000 input tokens, and 40,000 output tokens per 60-second window per organization, and production use with patient data requires accepting a BAA.

Glass is offering its clinical AI as a developer API. The core pitch: send clinical context to the Messages endpoint (POST https://glass.health/api/external/messages) and get back evidence-grounded answers with inline citations, differentials, and treatment options that your product can render in its own interface. The example request asks about first-line treatment for community-acquired pneumonia in a non-hospitalized adult, and the sample response distinguishes healthy adults (amoxicillin 1 g three times daily) from patients with comorbidities (combination therapy or a respiratory fluoroquinolone). Glass says it draws on 38M+ peer-reviewed citations, current guidelines, and 150,000+ FDA drug labels. There is also a separate Scribing API that takes encounter audio, returns a diarized transcript and a structured note in one call, supports 50+ languages and eight audio formats, and is billed at $0.85 per hour of audio. The commercial terms are the most concrete part of the page. Access is self-serve: sign up, get a key shown once in the sk-gla-api01 format, and pass it in an X-Api-Key header. Pricing starts at a $250 monthly minimum with usage counted against it, then per-token overage: $3 input and $16 output per million tokens for Glass 5.5, and $10 and $50 for Glass 5.0. Rate limits are 100 requests, 200,000 input tokens, and 40,000 output tokens per organization per 60-second window, with higher limits through sales. Every response includes exact token counts, which makes cost forecasting from real traffic straightforward. Production use with protected health information requires accepting a click-through BAA, though testing with synthetic or de-identified data can start immediately. The performance claims are thinner than the marketing suggests. Glass reports an error rate under 0.01%, multi-region deployment, and latency "optimized" for time-to-first-token, but gives no actual latency figures, uptime commitments, or independent benchmarks. The clinical claims are similarly unverified here: the page shows sample outputs and says every answer cites its sources, but offers no accuracy evaluation or guideline-concordance data. Anyone building clinical decision support on this should run their own validation on representative cases before relying on it, and should pay attention to the sample's own caveat that a patient with fever, neck stiffness, and light sensitivity needs same-day in-person evaluation rather than routine management.

Ask about this article