Health

Health models sized for a phone, scored on the exact file you ship.

Clinical & Medical

167 cases · 29 ranked

Numeric reasoning, extraction, contract-holding and safety, measured on the file you ship.

  1. 1=Gemma 4 E4B ITiPhone 17 Pro+89.8%
  2. 1=Nanbeige4.2 3BiPhone 17 Pro+86.2%
  3. 1=LFM2.5 2.6B86.1%
  4. 1=Spark X2.6 4BiPhone 17 Pro+86.0%
  5. 1=Ternary Bonsai 2 4B82.8%
  6. +1 tied

Wellness

160 cases · 29 ranked

Holds a wellness app's output contract: format, voice, language, the user's numbers.

  1. 1=Spark X2.6 4BiPhone 17 Pro+96.6%
  2. 1=Gemma 4 E4B ITiPhone 17 Pro+96.1%
  3. 1=Gemma 4 E2B ITiPhone 17 Pro+95.3%
  4. 1=Spark X2.5 4B94.3%
  5. 1=Granite 4.0 H Micro94.3%
  6. +3 tied

Agentic Clinical

140 cases · 29 ranked

Calls the right health tool, or none: asks instead of inventing, escalates red flags, carries doses exactly.

  1. 1=MedPsy 4BiPhone 17 Pro+94.4%
  2. 1=LFM2.5 2.6B91.1%
  3. 1=Gemma 4 E4B ITiPhone 17 Pro+89.3%
  4. 1=NVIDIA Nemotron 3 Nano 4B88.9%
  5. 1=Spark X2.6 4BiPhone 17 Pro+87.6%
  6. +1 tied

Multimodal Health

460 cases · 9 ranked

Reads synthetic health pictures and simulated phone photos, and knows when not to: reports, boxes, devices, charts, handwriting.

  1. 1=Gemma 4 E4B ITiPhone 17 Pro+73.8%
  2. 1=LFM3 VL 2B70.4%
  3. 1=Ministral 3 3B Instruct 2512iPhone 17 Pro+69.8%
  4. 1=Qwen3.5 2B67.1%
  5. 1=LFM2.5 VL 3B65.5%

System One

993 cases · 55 ranked · 18 phone builds

Models making bounded clinical decisions: each answer is one of the options supplied with the case.

  1. 1=Drex MaxLarger88.2%
  2. 1=mammoth-1Larger86.7%
  3. 1=d3-ultraLarger86.6%
  4. 2d1-ultraAPI only85.4%
  5. 4=d3-turboLarger84.3%
  6. +1 tied