NPCBench
Models follow patients with nasopharyngeal carcinoma through diagnosis and care. The main patient-level measure requires correct decisions at successive stages.
- Made by
- Pengkai Wang and colleagues
- Released
- November 2026
- Kind
- Medicine
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Capability
- Subject
- Clinical medicine
- Interaction
- Multi-turn
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Vision, Biology and medicine
- Strongest model tested
-
Gemini 3.1 Pro #40 on the capability index
- Last model added
- April 2026
- Scores refreshed
- November 20th, 2026
Claude Sonnet 4.6 leads with 23% among models released by April 2026 Claude Sonnet 4.6 leads with 23% among models released by April 2026
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper gives no overall score. The headline is strict patient success on CarePath episodes. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in April 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 23.0 Claude Sonnet 4.6
- 15.0 Gemini-2.5-Pro (snapshot unspecified)
- 11.0
Gemini 3.1 Pro
- 10.0 Hulu-Med-32B
- 9.00 Lingshu-7B
- 6.00
Gemini 2.5 Flash
- 6.00
Qwen3-VL-30B-A3B-Instruct
- 5.00
Qwen 3.6 35B-A3B
- 5.00
Qwen3.6 27B
- 2.00 Lingshu-32B
- 2.00 GPT-4.1
- 2.00
Qwen3-VL-8B-Instruct
- 2.00 GPT-5.1
- 1.00
Qwen3.5-27B
- 1.00
Qwen3.5-35B-A3B
- 1.00
Qwen3.5-9B
- 0.0 HealthGPT-Pro-4B
- 0.0 HealthGPT-Pro-8B
- 0.0 Hulu-Med-4B
- 0.0 Hulu-Med-7B
- 0.0
InternVL3.5-38B
- 1 Claude Sonnet 4.6 23%
- 2 Gemini-2.5-Pro (snapshot unspecified) 15%
- 3
Gemini 3.1 Pro 11%
- 4 Hulu-Med-32B 10%
- 5 Lingshu-7B 9.00%
- 6
Gemini 2.5 Flash Sep 2025 6.00%
- 7
Qwen3-VL-30B-A3B-Instruct 6.00%
- 8
Qwen 3.6 35B-A3B 5.00%
- 9
Qwen3.6 27B 5.00%
- 10 Lingshu-32B 2.00%
- 11 GPT-4.1 2.00%
- 12
Qwen3-VL-8B-Instruct 2.00%
- 13 GPT-5.1 2.00%
- 14
Qwen3.5-27B 1.00%
- 15
Qwen3.5-35B-A3B 1.00%
- 16
Qwen3.5-9B 1.00%
- 17 HealthGPT-Pro-4B 0%
- 18 HealthGPT-Pro-8B 0%
- 19 Hulu-Med-4B 0%
- 20 Hulu-Med-7B 0%
- 21
InternVL3.5-38B 0%
Source: The NPCBench paper
- 1 Claude Sonnet 4.6 23%
- 2
Gemini 3.1 Pro 11%
- 3
Gemini 2.5 Flash Sep 2025 6.00%
- 4
Qwen3-VL-30B-A3B-Instruct 6.00%
- 5
Qwen 3.6 35B-A3B 5.00%
- 6
Qwen3.6 27B 5.00%
- 7 GPT-4.1 2.00%
- 8
Qwen3-VL-8B-Instruct 2.00%
- 9 GPT-5.1 2.00%
- 10
Qwen3.5-27B 1.00%
- 11
Qwen3.5-35B-A3B 1.00%
- 12
Qwen3.5-9B 1.00%
- 13
InternVL3.5-38B 0%
Source: The NPCBench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 23% | Feb 2026 | Official leaderboard | |
| 2 | Gemini-2.5-Pro (snapshot unspecified) | 15% | Official leaderboard | ||
| 3 | | 11% | Feb 2026 | Official leaderboard | |
| 4 | Hulu-Med-32B | 10% | Official leaderboard | ||
| 5 | Lingshu-7B | 9.00% | Official leaderboard | ||
| 6 | | 6.00% | Sep 2025 | Official leaderboard | |
| 7 | | 6.00% | Oct 2025 | Official leaderboard | |
| 8 | | 5.00% | Apr 2026 | Official leaderboard | |
| 9 | | 5.00% | Apr 2026 | Official leaderboard | |
| 10 | Lingshu-32B | 2.00% | Official leaderboard | ||
| 11 | GPT-4.1 | 2.00% | Apr 2025 | Official leaderboard | |
| 12 | | 2.00% | Oct 2025 | Official leaderboard | |
| 13 | GPT-5.1 | 2.00% | Nov 2025 | Official leaderboard | |
| 14 | | 1.00% | Feb 2026 | Official leaderboard | |
| 15 | | 1.00% | Feb 2026 | Official leaderboard | |
| 16 | | 1.00% | Feb 2026 | Official leaderboard | |
| 17 | HealthGPT-Pro-4B | 0% | Official leaderboard | ||
| 18 | HealthGPT-Pro-8B | 0% | Official leaderboard | ||
| 19 | Hulu-Med-4B | 0% | Official leaderboard | ||
| 20 | Hulu-Med-7B | 0% | Official leaderboard | ||
| 21 | | 0% | Aug 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 23% | Feb 2026 | Official leaderboard | |
| 2 | | 11% | Feb 2026 | Official leaderboard | |
| 3 | | 6.00% | Sep 2025 | Official leaderboard | |
| 4 | | 6.00% | Oct 2025 | Official leaderboard | |
| 5 | | 5.00% | Apr 2026 | Official leaderboard | |
| 6 | | 5.00% | Apr 2026 | Official leaderboard | |
| 7 | GPT-4.1 | 2.00% | Apr 2025 | Official leaderboard | |
| 8 | | 2.00% | Oct 2025 | Official leaderboard | |
| 9 | GPT-5.1 | 2.00% | Nov 2025 | Official leaderboard | |
| 10 | | 1.00% | Feb 2026 | Official leaderboard | |
| 11 | | 1.00% | Feb 2026 | Official leaderboard | |
| 12 | | 1.00% | Feb 2026 | Official leaderboard | |
| 13 | | 0% | Aug 2025 | Official leaderboard |
Rankings by part
We read 13 parts from the benchmark’s paper, each ranking models on one part of the test, such as Guideline recall, Guideline application and Pathology. Gemini 3.1 Pro leads 8 of them, and 5 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Guideline recall
13 models we track have a score here.
- 1
Gemini 3.1 Pro ↑1 94%
- 2 Claude Sonnet 4.6 ↓1 93.2%
- 3 GPT-5.1 ↑6 90.4%
- 4 GPT-4.1 ↑3 90%
- 5
Qwen3.5-27B ↑5 89.2%
- 6
Gemini 2.5 Flash Sep 2025 ↓3 88.4%
- 7
InternVL3.5-38B ↑6 88%
- 8
Qwen3.6 27B ↓2 87.6%
- 9
Qwen3.5-35B-A3B ↑2 86.4%
- 10
Qwen3-VL-30B-A3B-Instruct ↓6 86%
Guideline application
13 models we track have a score here.
- 1
Gemini 3.1 Pro ↑1 92.5%
- 2 Claude Sonnet 4.6 ↓1 86.1%
- 3 GPT-5.1 ↑6 80.9%
- 4 GPT-4.1 ↑3 77.6%
- 5
Qwen3.5-27B ↑5 75.8%
- 6
Qwen3.5-35B-A3B ↑5 75.2%
- 7
Gemini 2.5 Flash Sep 2025 ↓4 71.8%
- 8
Qwen3.6 27B ↓2 71.4%
- 9
InternVL3.5-38B ↑4 69.7%
- 10
Qwen 3.6 35B-A3B ↓5 67.9%
Pathology
13 models we track have a score here.
- 1
Qwen3.5-9B ↑11 72.5%
- 2 GPT-4.1 ↑5 71%
- 3
Qwen3.5-27B ↑7 71%
- 4
Qwen3.6 27B ↑2 71%
- 5
Qwen3.5-35B-A3B ↑6 66.7%
- 6
Qwen 3.6 35B-A3B ↓1 66.7%
- 7
Gemini 2.5 Flash Sep 2025 ↓4 65.2%
- 8
Gemini 3.1 Pro ↓6 63.8%
- 9 GPT-5.1 62.3%
- 10
InternVL3.5-38B ↑3 59.4%
MRI T staging
13 models we track have a score here.
- 1
Qwen 3.6 35B-A3B ↑4 43.5%
- 2 GPT-4.1 ↑5 39.8%
- 3
Gemini 2.5 Flash Sep 2025 38%
- 4
Qwen3-VL-30B-A3B-Instruct 37.5%
- 5
Qwen3-VL-8B-Instruct ↑3 37.5%
- 6
Qwen3.6 27B 37.5%
- 7
Qwen3.5-27B ↑3 37%
- 8
InternVL3.5-38B ↑5 36.8%
- 9
Qwen3.5-35B-A3B ↑2 35.5%
- 10 Claude Sonnet 4.6 ↓9 33.5%
MRI N staging
13 models we track have a score here.
- 1
Qwen3.6 27B ↑5 42.8%
- 2
Gemini 3.1 Pro 38.5%
- 3 GPT-5.1 ↑6 32.8%
- 4 Claude Sonnet 4.6 ↓3 27.3%
- 5
Qwen3.5-27B ↑5 27.3%
- 6
Gemini 2.5 Flash Sep 2025 ↓3 25.8%
- 7 GPT-4.1 22%
- 8
Qwen3-VL-30B-A3B-Instruct ↓4 21.8%
- 9
Qwen3.5-35B-A3B ↑2 21.8%
- 10
Qwen3.5-9B ↑2 20.3%
MRI sequence
13 models we track have a score here.
- 1
Gemini 3.1 Pro ↑1 77%
- 2
Qwen3.6 27B ↑4 75.5%
- 3
Qwen3.5-27B ↑7 67%
- 4
Qwen3.5-9B ↑8 65.5%
- 5
Qwen3.5-35B-A3B ↑6 65%
- 6
Qwen 3.6 35B-A3B ↓1 61.5%
- 7
Gemini 2.5 Flash Sep 2025 ↓4 59%
- 8 Claude Sonnet 4.6 ↓7 57%
- 9 GPT-4.1 ↓2 48.5%
- 10 GPT-5.1 ↓1 47%
MRI structure
13 models we track have a score here.
- 1
Gemini 3.1 Pro ↑1 72%
- 2
Qwen3.5-35B-A3B ↑9 50.3%
- 3
Qwen3.5-27B ↑7 47.6%
- 4
Qwen 3.6 35B-A3B ↑1 44.8%
- 5
Qwen3.6 27B ↑1 41.4%
- 6
Qwen3.5-9B ↑6 38.9%
- 7
Qwen3-VL-30B-A3B-Instruct ↓3 36.2%
- 8 GPT-4.1 ↓1 28.3%
- 9 Claude Sonnet 4.6 ↓8 27.9%
- 10 GPT-5.1 ↓1 26.3%
Case planning
13 models we track have a score here.
- 1
Gemini 3.1 Pro ↑1 71.9%
- 2 Claude Sonnet 4.6 ↓1 69.2%
- 3 GPT-5.1 ↑6 63.5%
- 4
Qwen3.5-27B ↑6 61%
- 5
Gemini 2.5 Flash Sep 2025 ↓2 59.8%
- 6
Qwen3.5-9B ↑6 58%
- 7
Qwen3.5-35B-A3B ↑4 54.5%
- 8 GPT-4.1 ↓1 54.4%
- 9
Qwen3-VL-30B-A3B-Instruct ↓5 52.3%
- 10
Qwen3-VL-8B-Instruct ↓2 52.3%
Longitudinal revision per turn
13 models we track have a score here.
- 1
Gemini 3.1 Pro ↑1 87.9%
- 2 Claude Sonnet 4.6 ↓1 85.2%
- 3
Qwen3.6 27B ↑3 79.7%
- 4 GPT-4.1 ↑3 78.7%
- 5
Qwen3.5-27B ↑5 76.4%
- 6
Qwen 3.6 35B-A3B ↓1 76.1%
- 7 GPT-5.1 ↑2 75.7%
- 8
Qwen3.5-35B-A3B ↑3 72.8%
- 9
Qwen3-VL-8B-Instruct ↓1 69.8%
- 10
Gemini 2.5 Flash Sep 2025 ↓7 69.5%
Longitudinal revision all turns
13 models we track have a score here.
- 1
Gemini 3.1 Pro ↑1 71%
- 2 Claude Sonnet 4.6 ↓1 63%
- 3
Qwen3.6 27B ↑3 53%
- 4
Qwen 3.6 35B-A3B ↑1 49%
- 5
Qwen3.5-27B ↑5 46%
- 6 GPT-4.1 ↑1 45%
- 7 GPT-5.1 ↑2 45%
- 8
Gemini 2.5 Flash Sep 2025 ↓5 42%
- 9
Qwen3.5-35B-A3B ↑2 41%
- 10
Qwen3-VL-8B-Instruct ↓2 38%
CarePath care accuracy
13 models we track have a score here.
- 1
Gemini 3.1 Pro ↑1 89.9%
- 2 Claude Sonnet 4.6 ↓1 80.6%
- 3 GPT-5.1 ↑6 71%
- 4
Qwen3.6 27B ↑2 65.9%
- 5
Gemini 2.5 Flash Sep 2025 ↓2 65%
- 6
Qwen3.5-35B-A3B ↑5 63.1%
- 7
Qwen3.5-27B ↑3 62.7%
- 8 GPT-4.1 ↓1 61.8%
- 9
Qwen 3.6 35B-A3B ↓4 61.8%
- 10
Qwen3.5-9B ↑2 53%
CarePath strict success
13 models we track have a score here.
- 1 Claude Sonnet 4.6 23%
- 2
Gemini 3.1 Pro 11%
- 3
Gemini 2.5 Flash Sep 2025 6.00%
- 4
Qwen3-VL-30B-A3B-Instruct 6.00%
- 5
Qwen 3.6 35B-A3B 5.00%
- 6
Qwen3.6 27B 5.00%
- 7 GPT-4.1 2.00%
- 8
Qwen3-VL-8B-Instruct 2.00%
- 9 GPT-5.1 2.00%
- 10
Qwen3.5-27B 1.00%
Consultation rubric
Rubric score from 0 to 100. 13 models we track have a score here.
- 1 GPT-5.1 ↑8 58.4
- 2
Gemini 3.1 Pro 47.8
- 3 Claude Sonnet 4.6 ↓2 47.5
- 4
Qwen3.6 27B ↑2 47.2
- 5
Qwen3.5-27B ↑5 46.8
- 6
Qwen 3.6 35B-A3B ↓1 42.4
- 7
Qwen3.5-35B-A3B ↑4 41.4
- 8
Qwen3.5-9B ↑4 34.7
- 9
Gemini 2.5 Flash Sep 2025 ↓6 34.4
- 10 GPT-4.1 ↓3 33.3
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The NPCBench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official NPCBench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Law, finance and medicine
- MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy.
Qwen 3.8 Max 80.4%
- HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
- HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
- FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy.
Ling-3.1-flash 81.7%
- KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. o1-preview 92.7%
- Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for Qwen3.6 27B. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (21 entries).
- Pass Models scored 13 models we track, plus 8 other entries on its leaderboard.
- Warning Newest score The newest came 7 months ago, in April 2026, for Qwen3.6 27B.
- Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3.1 Pro (#40).
- Pass Headroom The best is 23%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 13 rows for models we track and 8 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.