Skip to content
← AI by field
Clinicians, hospitals and health companies

AI for medicine and health, right now

Answering clinical questions, reading scans and records, drafting notes and supporting diagnosis.

Moving: fastUpdated Nov 12, 2026 from 11 sources
Open medicine in Spezia Free account. No card.
Where AI is in medicine now

AI models now match or beat physicians on some diagnostic benchmarks, but real-world clinical reliability and regulatory approval remain limited.

What models can do now

  • Microsoft's MAI-DxO system diagnosed complex cases more than twice as accurately as physicians with textbooks and internet, and achieved 80% accuracy on typical emergency room cases versus 47% for doctors.[8]
  • Claude Sonnet 5.5 leads HealthBench Professional with 69.2%, up from 38.4% five months earlier.[1]
  • AI-derived ejection fraction measurements from cardiac MRI predicted patient mortality better than clinician measurements and took 2-3 seconds per scan versus 13 minutes for experts.[9]
  • An AI system improved emergency physicians' sensitivity for chest X-ray anomaly detection from 49.2% to 78.8%.[10]
  • A neural network predicted obstructive sleep apnea severity from 18 clinical parameters with AUC values of 0.89 to 0.96.[4]
  • Open-source prostate biopsy classifiers achieved mean AUC of 0.982 for cancer detection and 0.949 for high-grade cancer on an independent dataset.[11]

Where they still fall short

  • Big Pharma CEOs say AI cannot yet predict human biology or safety, and no AI-designed drug has received FDA approval.[6]
  • GPT-4 produced 82 errors in breast cancer clinical note augmentations, 28 of which were potentially harmful or misleading.[7]
  • Top LLMs score poorly on chemistry reasoning, with Gemini 3 Pro achieving only 27.8% scope and 10.7% depth on ChemInsight.[5]

What is expected next

  • Harvard Medical School will test whether an ambient AI coach improves physicians' leadership in simulated trauma resuscitations.[2]
  • Australia plans mandatory AI safety rules and a regulator after an OpenAI agent breached a Medicare portal.[3]

How fast it is moving

The record on this field's benchmarks over the last year, from Spezia's model leaderboard.

HealthBench Professional

Conversations in the style of real clinicians' chats with AI, graded against rubrics written by physicians.

Record: 69.2%, up from 38.4% 5 months ago (1.8×) (Claude Sonnet 5.5)

0%20%40%60%80%Claude Sonnet 5.5Claude Fable 5
Jul 26Oct 26

the record so far set a record other models

HealthBench

5,000 realistic health conversations with patients and clinicians, each graded against criteria written by 262 physicians.

Record: 65.4%, up from 58% a year ago (Claude Sonnet 5.5)

0%20%40%60%80%Claude Sonnet 5.5Kimi K2-Thinking-0905
Oct 25Jan 26Apr 26Jul 26Oct 26

the record so far set a record other models

The best models for medicine now

On this field's benchmarks alone.

  1. 1Claude Sonnet 5.5
  2. 2Claude Fable 5.2
  3. 3Claude Haiku 5.5
  4. 4GPT-6 Astra
  5. 5Claude Opus 5.5
  6. 6GPT-6.1 Sol

Milestones

What became possible, newest first, each marked by how firm it is. Open one for the words it rests on.

Showndemonstrated with published results, a paper or an independent checkClaimedsaid by the company or people who built it, not yet checked by othersReportedreported by others, not confirmed by the company or a published resultNot publicdone by a model that is not released to the public

  1. Nov 9, 2026ShownMAI-DxO
    Microsoft's MAI-DxO system outperformed physicians on complex diagnostic cases, achieving more than twice the accuracy.
    “MAI-DxO was more than twice as accurate as physicians armed with textbooks and the internet” Microsoft’s Medical AI Is Over Twice as Good Diagnosing Diseases as Doctors
  2. Nov 9, 2026ShownMAI-DxO
    Microsoft's MAI-DxO system diagnosed complex cases more than twice as accurately as physicians with textbooks and internet, and achieved 80% accuracy on typical emergency room cases versus 47% for doctors.
    “MAI-DxO was more than twice as accurate as physicians armed with textbooks and the internet, and nearly four times as accurate as those relying solely on their own knowledge.” Microsoft’s Medical AI Is Over Twice as Good Diagnosing Diseases as Doctors
  3. Nov 8, 2026Shown
    A neural network predicted obstructive sleep apnea severity with AUC values from 0.89 to 0.96.
    “The ANN model demonstrated good discriminative performance across OSA severity categories, with class-specific AUC values ranging from 0.89 to 0.96.” Neural network predicts obstructive sleep apnea severity from 18 clinical parame
  4. Nov 6, 2026ShownISM6331
    Insilico's generative AI-designed TEAD inhibitor ISM6331 showed a 40% response rate in Phase 1 mesothelioma patients.
    “ISM6331 produced a 40.0% response rate at 130 mg once daily, though that figure rests on two of five evaluable patients” Insilico's generative AI-designed TEAD inhibitor ISM6331 showed a 40% response r
  5. Nov 6, 2026Shown
    A federated, quantum-resistant blockchain framework for breast cancer diagnosis achieved 94.23% accuracy.
    “The proposed centralized model achieved an accuracy of 94.23%, precision of 94.00%, recall of 94.47%, F1-score of 94.24%, and ROC-AUC of 0.9651.” Researchers propose a federated, quantum-resistant blockchain framework for priv
  6. Nov 4, 2026Shown
    Multimodal deep learning combining skin images and clinical metadata improved skin lesion classification accuracy over baselines.
    “integrating structured clinical metadata with skin images was associated with improved skin lesion classification performance relative to comparator models” Multimodal deep learning combining skin images and clinical metadata improved sk
Every milestone, story and model Pick a model to see what it can do in medicine.

Sources

  1. [1]HealthBench Professional: best result 69.2%, up from 38.4% 5 months ago (1.8×)Leaderboard · 2026-09-28
  2. [2]Harvard Medical School will test whether an ambient AI coach improves physicians' leadership in simulated trauma resuscitations.Spezia news · 2026-11-09
  3. [3]Australia plans mandatory AI safety rules and a regulator after an OpenAI agent breached a Medicare portal.Spezia news · 2026-11-09
  4. [4]Neural network predicts obstructive sleep apnea severity from 18 clinical parameters with AUC of 0.89 to 0.96.Spezia news · 2026-11-08
  5. [5]New ChemInsight benchmark finds top LLMs score poorly on chemistry reasoning, with Gemini 3 Pro leading.Spezia news · 2026-11-08
  6. [6]Big Pharma CEOs say AI cannot yet predict human biology or safety, and no AI-designed drug is FDA-approved.Spezia news · 2026-11-08
  7. [7]Study finds GPT-4 breast cancer note augmentations produced 82 errors, 28 potentially harmful or misleading.Spezia news · 2026-11-04
  8. [8]Microsoft’s Medical AI Is Over Twice as Good Diagnosing Diseases as DoctorsWeb · 2026-11-03
  9. [9]AI as first-line tool for analysing heart scans?Web · 2026-11-10
  10. [10]Cross-sectional study evaluating the impact of an artificial intelligence-assisted system on diagnostic accuracy for chest and bone X-rays in the emergency department - Scandinavian Journal of Trauma, Resuscitation and Emergency MedicineWeb · 2026-11-11
  11. [11]294 Performance of Open-source–derived AI Algorithms for Prostate Biopsy Image Classification on an Independent DatasetWeb · 2026-10-31

Other fields