Local model reads a 131K-token patient history in 61s, then reasons for 11 minutes
maniac518 · reddit · 2026-11-19
Developer SudoingX stress-tested AntLing's Ling-3.0-flash-Sante medical model with a synthetic patient's entire record — 334 visits across 71 years, 130,993 tokens in a single prompt — asking for current medications, allergies, and active conditions with dates.
- Hardware: two DGX Sparks with official FP8 weights, 256GB unified memory; reading the full record took 61 seconds (2,148 tok/s)
- Reasoning: the model thought for 30,865 tokens (11 minutes) — roughly 11x the read time
- Score: 7/7 allergies, 8/10 active conditions, 2/7 dental codes, 2/5 medications; every stopped item stayed off the active lists, and it caught a medication started in 1954 still running 67 years later
The author stresses it's a research model, not a medical device, and med lists need verification. The hard part is records full of start/stop/restart items where 'current' hinges on getting each date right.
More from Infra
- Micron's Scott DeBoer says 3D DRAM is coming in 2028/2029 — firstadopter · 2026-11-19
- Two Energy Clocks for AI Infrastructure: Flexibility Buys Time, Capacity Sets Scale — ingliguori · 2026-11-19
- NVIDIA to demo Cosmos Data Factory pipeline for training-ready synthetic datasets with Voxel51 and Encord — NVIDIAAI · 2026-11-19
- Analyst: memory has shifted from commodity to value, explaining sustainably higher margins — BenBajarin · 2026-11-19
- Microsoft open-sources Aion 2.3B on-device SLM, shipping as the Windows built-in model — jacek2023 · 2026-11-19
- Voters rejected 46 of 47 local data center ballot measures, analysis finds — WillRinehart · 2026-11-19