FHIR-AgentBench (Single-Turn + FHIR Query Generator): leaderboard
Metric: Answer correctness (%; o4-mini judge of the free-text answer against the SQL-derived answer, 2,931 questions; one turn: the LLM writes FHIR API queries). Source: arxiv.org. Saturation forecast: Around 2029. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Flash | 32 |
| 2 | Qwen 3 32B | 30 |
| 3 | Llama 3.3 70B | 27 |
| 4 | O4 Mini | 25 |
Interactive version: theaggregate.ai/benchmark?slug=fhir-agentbench-single-turn-plus-fhir-query-generator · How It Works · Data refreshed daily, snapshot 2026-09-26.