Corti
The most complete healthcare AI stack you can buy as an API, and the only clinical vendor that speaks both MCP and A2A properly. Check the eval and fine-tuning gaps before you build your product on it.
PRICING
| Free credits | $50 | no card required, credits valid one year |
| Speech to Text | $0.0065/min | medical ASR |
| Medical Coding / Agents / Text Generation | $4 in / $16 out per mTok | one credit balance across all APIs |
| corti-s1 | $1 in / $8 out per mTok | $0.20/mTok cached input |
| corti-s1-mini | $1 in / $4 out per mTok | $0.10/mTok cached input |
| Acceleration Pack | $1,000/mo | 1,000 credits plus 5 developer support hours per month |
| Enterprise | Custom | custom-trained models, sovereign cloud, priority throughput |
Corti bills one credit balance usable across every API in dev and production. EUR figures are converted from the published USD list prices and will move with the rate. Corti Advance is a rolling monthly commitment with no published price.
checked 2026-08-18 · vendor pricing page
Element scores
Strengths
Corti covers 19 of the 20 elements, which no other clinical AI vendor in this database comes close to. The governance layer is the real product: a deterministic orchestrator validates every agent action against clinical protocols before it executes, and every decision, tool call and handoff is recorded as a replayable trace. Protocol support is best-in-class, with full MCP and A2A compatibility so custom agents inherit the same governance. Speech-to-text is measurably ahead of the legacy incumbent at 4.6% WER against Dragon Medical One's 5.7%. EU sovereign hosting with no US cloud dependency is a genuine differentiator for European buyers, and open-sourcing GIM, currently top of the Hugging Face interpretability benchmark, is more transparency than most frontier labs ship.
Honest dings
The eval story is thin for a platform selling into regulated deployment: customers get a feedback endpoint, not a regression harness, and Quality Assurance is still marked 'coming soon' on half the text-generation endpoints. There is no hosted vector store and no self-serve fine-tuning, so adaptation means an Enterprise conversation. The Agent SDK, the thing the whole agentic pitch rests on, is alpha v2 private preview with a Python client that only implements agents. The HealthBench Professional win is Corti running its own multi-agent harness against base models, which is a fair engineering result but not the like-for-like model comparison the chart implies. Separately, Seattle Fire Department ran Corti's AI on every 911 medical call from December 2023 without public disclosure or review under the city's surveillance ordinance, and could not produce metrics on how the system was performing. That is a governance failure by the customer, not a product defect, but it is the deployment pattern this product enables and buyers should plan disclosure before go-live.
Sources (16) — every claim traceable
Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).
- corti.ai/pricing — Official list pricing: $50 free credits, STT $0.0065/min, $4/$16 per mTok on coding/agents/textgen, s1 and s1-mini tiers, Acceleration Pack $1,000/mo, credits valid one year (accessed 2026-08-18)
- docs.corti.ai/models/welcome.md — Corti Models: EU sovereign hosting, OpenAI-compatible API, embedding models, s1 family in frontier and efficient tiers with and without reasoning (accessed 2026-08-18)
- docs.corti.ai/agentic/overview.md — Agentic Framework: guardrails, replayable traces, persistent thread state, FHIR runtime context, human-approval pause/resume, prebuilt Experts (accessed 2026-08-18)
- docs.corti.ai/agentic/agent-sdk/concepts.md — Agent SDK is alpha v2 private preview; TypeScript exposes agents/contexts/registry/usage/feedback clients, Python ships AgentsClient only (accessed 2026-08-18)
- docs.corti.ai/textgen/overview.md — FactsR and Guided Documents, schema-controlled typed outputs, versioned sections and templates; Quality Assurance marked 'coming soon' on Guided Documents and Streams (accessed 2026-08-18)
- corti.ai/guides/a2a-and-mcp-how-the-corti-agentic-f… — A2A for agent-to-agent and MCP for agent-to-tool; all pre-built Experts MCP-compatible, customer MCP servers callable by the orchestrator (accessed 2026-08-18)
- corti.ai/news/corti-launches-agentic-infrastructure… — Official launch: deterministic guardrails validating actions in sub-milliseconds, end-to-end audit trails, MCP and A2A support, production Agent Library (accessed 2026-08-18)
- corti.ai/benchmarks — Corti's own HealthBench Professional results: Symphony 60.5% overall vs GPT-5.4 48.1%, red-teaming 59% vs 30.3% (vendor-run) (accessed 2026-08-18)
- venturebeat.com/technology/cortis-new-symphony-for-… — Independent reporting on speech-to-text benchmarks: 4.6% WER vs Dragon Medical One 5.7%, 93.5% vs 92.9% medical term recall, 98.3% formatted entity recall (accessed 2026-08-18)
- prnewswire.com/news-releases/corti-steps-up-its-sup… — Official company statement 13/05/2026: over 100 million patients annually, health systems including the NHS, offices Copenhagen/New York/London (accessed 2026-08-18)
- cnbc.com/2025/12/30/corti-will-go-public-but-not-in… — CEO Andreas Cleve: 250,000 patient interactions per day, $60M Series B at $260M valuation led by Prosus Ventures and Atomico, no 2026 IPO (accessed 2026-08-18)
- cbinsights.com/company/corti — Independent company profile: founded 2016, Copenhagen HQ, $88.86M total raised, Series B stage (accessed 2026-08-18)
- geekwire.com/2026/report-seattle-using-ai-to-route-… — Seattle Fire Department used Corti on all 911 medical calls from December 2023 with no public disclosure and no review under the city surveillance ordinance (accessed 2026-08-18)
- firerescue1.com/artificial-intelligence/report-find… — Seattle Times detail: ~$260,000 annual contract, department has no public performance metrics, Corti barred from training other customers' models on Seattle data (accessed 2026-08-18)
- corti.ai/stories/gim-a-new-standard-for-mechanistic… — GIM (Gradient Interaction Modifications) open-sourced as a Python package, top-ranked on the Hugging Face Mechanistic Interpretability Benchmark (accessed 2026-08-18)
- corti.ai/agents — Agent Library and governance FAQ: central orchestrator validates each action, agents cannot act outside defined scope, customisable system prompts and Experts (accessed 2026-08-18)