Med-HALT
Models answer medical exam questions designed to reveal hallucinations. The hub port uses the false-confidence, fake-question, and none-of-the-above reasoning tasks.
On October 14, 2023, Llama 2 70B led the reasoning table at 72.33% average accuracy. GPT-3.5 scored 44.48%.
- Made by
- Med-HALT authors
- Released
- July 2023
- Kind
- Hallucination rates
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Propensity
- Risk
- Specialized advice
- Behavior
- Hallucination
- Subject
- Clinical medicine
- Language
- English
- Interaction
- Single turn
- Items from
- Exams and contests
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Biology and medicine, Content safety
- Strongest model tested
-
Llama 2-70B #210 on the capability index
- Last model added
- July 2023
- Scores refreshed
- November 20th, 2026
Falcon-40B leads with 42.7% among models released by July 2023 Falcon-40B leads with 42.7% among models released by July 2023
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The headline is the unweighted mean of accuracy across the three reasoning tests and four memory tests. 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 July 2023, 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
- 1
Falcon-40B 42.7%
- 2 Falcon 40B-instruct 39.7%
- 3
Llama 2-70B 35.6%
- 4 davinci-003 34.6%
- 5 GPT-3.5 (snapshot unspecified) 30.5%
- 6
Llama 2-13B 29.3%
- 7
MPT-7B 24.3%
- 8 Llama-2-13B-chat 20.4%
- 9
Llama 2-7B 18.9%
- 10 Mpt 7B instruct 18%
- 11 Llama-2-7B-chat 13.9%
- 12 Llama-2 70B Chat 12.3%
Source: The Med-HALT paper
- 1
Falcon-40B 42.7%
- 2
Llama 2-70B 35.6%
- 3 davinci-003 34.6%
- 4
Llama 2-13B 29.3%
- 5
MPT-7B 24.3%
- 6
Llama 2-7B 18.9%
Source: The Med-HALT paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 42.7% | Mar 2023 | Official leaderboard | |
| 2 | Falcon 40B-instruct | 39.7% | Official leaderboard | ||
| 3 | | 35.6% | Jul 2023 | Official leaderboard | |
| 4 | davinci-003 | 34.6% | Nov 2022 | Official leaderboard | |
| 5 | GPT-3.5 (snapshot unspecified) | 30.5% | Official leaderboard | ||
| 6 | | 29.3% | Jul 2023 | Official leaderboard | |
| 7 | | 24.3% | May 2023 | Official leaderboard | |
| 8 | Llama-2-13B-chat | 20.4% | Official leaderboard | ||
| 9 | | 18.9% | Jul 2023 | Official leaderboard | |
| 10 | Mpt 7B instruct | 18% | Official leaderboard | ||
| 11 | Llama-2-7B-chat | 13.9% | Official leaderboard | ||
| 12 | Llama-2 70B Chat | 12.3% | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 42.7% | Mar 2023 | Official leaderboard | |
| 2 | | 35.6% | Jul 2023 | Official leaderboard | |
| 3 | davinci-003 | 34.6% | Nov 2022 | Official leaderboard | |
| 4 | | 29.3% | Jul 2023 | Official leaderboard | |
| 5 | | 24.3% | May 2023 | Official leaderboard | |
| 6 | | 18.9% | Jul 2023 | Official leaderboard |
Rankings by part
We read 7 parts from the benchmark’s paper in 2 groups: reasoning and memory. Each ranks models on one part of the test. Falcon-40B leads 4 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.
Reasoning FCT
6 models we track have a score here.
- 1
Llama 2-70B ↑1 42.2%
- 2
Falcon-40B ↓1 18.7%
- 3 davinci-003 16.8%
- 4
Llama 2-13B 1.72%
- 5
MPT-7B 0.85%
- 6
Llama 2-7B 0.45%
Reasoning Fake
6 models we track have a score here.
- 1
Falcon-40B 99.9%
- 2
Llama 2-70B 97.3%
- 3
Llama 2-13B ↑1 89.5%
- 4 davinci-003 ↓1 82.7%
- 5
Llama 2-7B ↑1 58.7%
- 6
MPT-7B ↓1 48.5%
Reasoning Nota
6 models we track have a score here.
- 1
Llama 2-70B ↑1 77.5%
- 2
Llama 2-13B ↑2 74.4%
- 3
Llama 2-7B ↑3 69.5%
- 4 davinci-003 ↓1 63.9%
- 5
Falcon-40B ↓4 58.7%
- 6
MPT-7B ↓1 19.9%
IR Pmid2Title
6 models we track have a score here.
- 1
Falcon-40B 40.5%
- 2
MPT-7B ↑3 20.1%
- 3
Llama 2-13B ↑1 0.53%
- 4
Llama 2-70B ↓2 0.12%
- 5 davinci-003 ↓2 0.02%
- 6
Llama 2-7B 0%
IR Title2Pubmedlink
6 models we track have a score here.
- 1
Falcon-40B 40.5%
- 2
MPT-7B ↑3 40.5%
- 3 davinci-003 38.5%
- 4
Llama 2-70B ↓2 14.8%
- 5
Llama 2-13B ↓1 10.6%
- 6
Llama 2-7B 3.72%
IR Abstract2Pubmedlink
6 models we track have a score here.
- 1
Falcon-40B 40.5%
- 2 davinci-003 ↑1 40.4%
- 3
MPT-7B ↑2 40%
- 4
Llama 2-70B ↓2 17.2%
- 5
Llama 2-13B ↓1 4.70%
- 6
Llama 2-7B 0.26%
IR Pubmedlink2Title
6 models we track have a score here.
- 1
Llama 2-13B ↑3 23.7%
- 2
Falcon-40B ↓1 0.06%
- 3
Llama 2-70B ↓1 0.02%
- 4 davinci-003 ↓1 0%
- 5
MPT-7B 0%
- 6
Llama 2-7B 0%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The Med-HALT 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 Med-HALT 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 Factuality
- SimpleVQA In use Models answer 2,025 short factual questions about images across 9 tasks and 9 topics. An LLM judge grades answers against fixed references.
Qwen3.7-Plus 81.7%
- HallusionBench In use Models answer 1,129 yes or no questions about 346 images, built in pairs to expose language hallucination and visual illusion. Scored by accuracy.
Keye-VL-2.0-30B-A3B 73.5%
- SimpleQA Verified In use 1,000 human-verified, de-duplicated short-answer factual questions measuring parametric knowledge.
Gemini 3.1 Pro 77.5%
- AA-Omniscience v1.1 In use Models answer 3,000 knowledge questions in six domains without tools and may abstain. The index adds correct answers and subtracts wrong ones.
Inkling 2 Small 16.1
- GaRAGe In use Models write long-form RAG answers to 2,366 questions with human-annotated grounding passages. Scoring checks whether answers use the relevant passages.
Amazon Nova Pro 60.7%
- Kimi Hallu Bench In use Models answer hard prompts drawn from practical work tasks. Fact-checking agents score the share of claims that are hallucinated. GPT-6 Astra 4.20%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 3 years ago, in July 2023, for Llama 2-13B. 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 6 models we track, plus 6 other entries on its leaderboard.
- Warning Newest score The newest came 3 years ago, in July 2023, for Llama 2-13B.
- Note Frontier models None of the top 10 has a score. The strongest tested is Llama 2-70B (#210).
- Pass Headroom The best is 42.7%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on aclanthology.org: 6 rows for models we track and 6 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.