CareQA
Models answer questions from Spanish specialist healthcare examinations. The evaluation includes English and Spanish versions and an open-response format.
- Made by
- Anna Arias-Duart and colleagues
- Released
- February 2025
- Kind
- Medicine
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Capability
- Subject
- Clinical medicine
- Interaction
- Single turn
- Items from
- Exams and contests, Translated
- Refresh
- One fixed set
- Run by
- Anyone
- Occupations
- Healthcare practitioners and technical
- In views
- Biology and medicine
- Strongest model tested
-
phi-3-medium 14B #194 on the capability index
- Last model added
- June 2024
- Scores refreshed
- November 20th, 2026
Meta-Llama-3.1-70B-Instruct tops the leaderboard with 83.7% phi-3-medium 14B leads with 76.9% among models released by June 2024
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 reports no overall score, so this is the plain mean of its ten close-ended accuracy parts. 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. The best model we track is phi-3-medium 14B, at 76.9%. This view shows only models we track, each at its best.
The newest model with a score here came out in June 2024, 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
- 83.7 Meta-Llama-3.1-70B-Instruct
- 76.9
phi-3-medium 14B
- 73.3 Yi-1.5-34B-Chat
- 72.5 Mixtral-8x7B-Instruct-v0.1
- 70.1
phi-3-mini 3.8B
- 70.0 Meta-Llama-3.1-8B-Instruct
- 68.3 Llama3-Med42-8B
- 68.0
Qwen2-7B-Instruct
- 67.2 JSL-MedLlama-3-8B-v2.0
- 65.0
Yi-1.5-9B-Chat
- 62.9 BioMistral-MedMNX
- 60.7 Mistral-7B-Instruct-v0.3
- 1 Meta-Llama-3.1-70B-Instruct 83.7%
- 2
phi-3-medium 14B 76.9%
- 3 Yi-1.5-34B-Chat 73.3%
- 4 Mixtral-8x7B-Instruct-v0.1 72.5%
- 5
phi-3-mini 3.8B 70.1%
- 6 Meta-Llama-3.1-8B-Instruct 70%
- 7 Llama3-Med42-8B 68.3%
- 8
Qwen2-7B-Instruct 68%
- 9 JSL-MedLlama-3-8B-v2.0 67.2%
- 10
Yi-1.5-9B-Chat 65%
- 11 BioMistral-MedMNX 62.9%
- 12 Mistral-7B-Instruct-v0.3 60.7%
Source: The CareQA paper
- 1
phi-3-medium 14B 76.9%
- 2
phi-3-mini 3.8B 70.1%
- 3
Qwen2-7B-Instruct 68%
- 4
Yi-1.5-9B-Chat 65%
Source: The CareQA paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Meta-Llama-3.1-70B-Instruct | 83.7% | Official leaderboard | ||
| 2 | | 76.9% | Apr 2024 | Official leaderboard | |
| 3 | Yi-1.5-34B-Chat | 73.3% | Official leaderboard | ||
| 4 | Mixtral-8x7B-Instruct-v0.1 | 72.5% | Official leaderboard | ||
| 5 | | 70.1% | Apr 2024 | Official leaderboard | |
| 6 | Meta-Llama-3.1-8B-Instruct | 70% | Official leaderboard | ||
| 7 | Llama3-Med42-8B | 68.3% | Official leaderboard | ||
| 8 | | 68% | Jun 2024 | Official leaderboard | |
| 9 | JSL-MedLlama-3-8B-v2.0 | 67.2% | Official leaderboard | ||
| 10 | | 65% | May 2024 | Official leaderboard | |
| 11 | BioMistral-MedMNX | 62.9% | Official leaderboard | ||
| 12 | Mistral-7B-Instruct-v0.3 | 60.7% | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 76.9% | Apr 2024 | Official leaderboard | |
| 2 | | 70.1% | Apr 2024 | Official leaderboard | |
| 3 | | 68% | Jun 2024 | Official leaderboard | |
| 4 | | 65% | May 2024 | 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 English accuracy, Spanish accuracy and Bits per Byte. phi-3-medium 14B leads 9 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
English accuracy
4 models we track have a score here.
- 1
phi-3-medium 14B 76.9%
- 2
phi-3-mini 3.8B 70.1%
- 3
Qwen2-7B-Instruct 68%
- 4
Yi-1.5-9B-Chat 65%
Spanish accuracy
4 models we track have a score here.
- 1
phi-3-medium 14B 71.8%
- 2
Qwen2-7B-Instruct ↑1 62.1%
- 3
phi-3-mini 3.8B ↓1 58.5%
- 4
Yi-1.5-9B-Chat 50.7%
Bits per Byte
4 models we track have a score here.
- 1
phi-3-medium 14B 1.25
- 2
phi-3-mini 3.8B 1.34
- 3
Qwen2-7B-Instruct 1.47
- 4
Yi-1.5-9B-Chat 1.54
Byte Perplexity
4 models we track have a score here.
- 1
phi-3-medium 14B 2.39
- 2
phi-3-mini 3.8B 2.54
- 3
Qwen2-7B-Instruct 2.77
- 4
Yi-1.5-9B-Chat 2.90
Word Perplexity
4 models we track have a score here.
- 1
phi-3-medium 14B 375.4
- 2
phi-3-mini 3.8B 566.1
- 3
Qwen2-7B-Instruct 1024.4
- 4
Yi-1.5-9B-Chat 1416.8
Prometheus
4 models we track have a score here.
- 1
Yi-1.5-9B-Chat ↑3 0.59
- 2
Qwen2-7B-Instruct ↑1 0.46
- 3
phi-3-medium 14B ↓2 0.32
- 4
phi-3-mini 3.8B ↓2 0.18
BERTScore
4 models we track have a score here.
- 1
Yi-1.5-9B-Chat ↑3 0.83
- 2
phi-3-medium 14B ↓1 0.81
- 3
Qwen2-7B-Instruct 0.76
- 4
phi-3-mini 3.8B ↓2 0.68
BLEU
4 models we track have a score here.
- 1
phi-3-medium 14B 0.01
- 2
Yi-1.5-9B-Chat ↑2 0.00
- 3
phi-3-mini 3.8B ↓1 0.00
- 4
Qwen2-7B-Instruct ↓1 0.00
BLEURT
4 models we track have a score here.
- 1
Yi-1.5-9B-Chat ↑3 -1.18
- 2
Qwen2-7B-Instruct ↑1 -1.23
- 3
phi-3-medium 14B ↓2 -1.28
- 4
phi-3-mini 3.8B ↓2 -1.28
MoverScore
4 models we track have a score here.
- 1
phi-3-mini 3.8B ↑1 0.50
- 2
phi-3-medium 14B ↓1 0.50
- 3
Qwen2-7B-Instruct 0.50
- 4
Yi-1.5-9B-Chat 0.49
ROUGE1
4 models we track have a score here.
- 1
phi-3-medium 14B 0.09
- 2
Yi-1.5-9B-Chat ↑2 0.08
- 3
Qwen2-7B-Instruct 0.07
- 4
phi-3-mini 3.8B ↓2 0.06
ROUGE2
4 models we track have a score here.
- 1
phi-3-medium 14B 0.03
- 2
Yi-1.5-9B-Chat ↑2 0.02
- 3
Qwen2-7B-Instruct 0.02
- 4
phi-3-mini 3.8B ↓2 0.02
ROUGEL
4 models we track have a score here.
- 1
phi-3-medium 14B 0.08
- 2
Yi-1.5-9B-Chat ↑2 0.07
- 3
Qwen2-7B-Instruct 0.06
- 4
phi-3-mini 3.8B ↓2 0.05
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The CareQA 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 CareQA 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 2 years ago, in June 2024, for Qwen2-7B-Instruct. 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 (12 entries).
- Pass Models scored 4 models we track, plus 8 other entries on its leaderboard.
- Warning Newest score The newest came 2 years ago, in June 2024, for Qwen2-7B-Instruct.
- Note Frontier models None of the top 10 has a score. The strongest tested is phi-3-medium 14B (#194).
- Pass Headroom The best is 76.9%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 4 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.