HalluDial
Models detect hallucinations in dialogue.
The publisher reports model results.
- Made by
- BAAI FlagEval
- Released
- June 2024
- Kind
- Hallucination rates
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Generated
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- GPT-4o May 2024 #162 on the capability index
- Last model added
- May 2024
- Scores refreshed
- November 20th, 2026
HalluJudge tops the leaderboard with 84.9 GPT-4 Turbo (Jan 2024) leads with 80.2 among models released by May 2024
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. 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 GPT-4 Turbo (Jan 2024), at 80.2. This view shows only models we track, each at its best.
The newest model with a score here came out in May 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
- 84.9 HalluJudge
- 80.2 GPT-4 Turbo (Jan 2024)
- 78.8 GPT-4o
- 76.4 GPT-4 Turbo
- 67.9 GPT-3.5 Turbo
- 66.3
DeepSeek LLM 67B
- 66.2 Qwen1.5-32B-Chat
- 65.9 Baichuan2-13B-Chat
- 65.4
Qwen1.5-7B-Chat
- 65.2
Qwen1.5-72B
- 64.5 internlm2-chat-20B
- 61.8 internlm2-chat-7B
- 60.1 Llama-2-70B-chat
- 58.9 vicuna-33B-v1.3
- 57.3 Baichuan2-7B-Chat
- 54.8 vicuna-13B-v1.5
- 53.4 ChatGLM3-6B
- 47.9 vicuna-7B-v1.5
- 45.9 deepseek-llm-7B-chat
- 34.5 Llama-2-13B-chat
- 28.6 Llama-2-7B-chat
- 1 HalluJudge 84.9
- 2 GPT-4 Turbo (Jan 2024) 80.2
- 3 GPT-4o May 2024 78.8
- 4 GPT-4 Turbo Nov 2023 76.4
- 5 GPT-3.5 Turbo Jan 2024 67.9
- 6
DeepSeek LLM 67B 66.3
- 7 Qwen1.5-32B-Chat 66.2
- 8 Baichuan2-13B-Chat 65.9
- 9
Qwen1.5-7B-Chat 65.4
- 10
Qwen1.5-72B 65.2
- 11 internlm2-chat-20B 64.5
- 12 internlm2-chat-7B 61.8
- 13 Llama-2-70B-chat 60.1
- 14 vicuna-33B-v1.3 58.9
- 15 Baichuan2-7B-Chat 57.3
- 16 vicuna-13B-v1.5 54.8
- 17 ChatGLM3-6B 53.4
- 18 vicuna-7B-v1.5 47.9
- 19 deepseek-llm-7B-chat 45.9
- 20 Llama-2-13B-chat 34.5
- 21 Llama-2-7B-chat 28.6
Source: The HalluDial paper
- 1 GPT-4 Turbo (Jan 2024) 80.2
- 2 GPT-4o May 2024 78.8
- 3 GPT-4 Turbo Nov 2023 76.4
- 4 GPT-3.5 Turbo Jan 2024 67.9
- 5
DeepSeek LLM 67B 66.3
- 6 Baichuan2-13B-Chat 65.9
- 7
Qwen1.5-7B-Chat 65.4
- 8
Qwen1.5-72B 65.2
- 9 ChatGLM3-6B 53.4
Source: The HalluDial paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | HalluJudge | 84.9 | Official leaderboard | ||
| 2 | GPT-4 Turbo (Jan 2024) | 80.2 | Jan 2024 | Official leaderboard | |
| 3 | GPT-4o May 2024 | 78.8 | May 2024 | Official leaderboard | |
| 4 | GPT-4 Turbo Nov 2023 | 76.4 | Nov 2023 | Official leaderboard | |
| 5 | GPT-3.5 Turbo Jan 2024 | 67.9 | Jan 2024 | Official leaderboard | |
| 6 | | 66.3 | Nov 2023 | Official leaderboard | |
| 7 | Qwen1.5-32B-Chat | 66.2 | Official leaderboard | ||
| 8 | Baichuan2-13B-Chat | 65.9 | Sep 2023 | Official leaderboard | |
| 9 | | 65.4 | Feb 2024 | Official leaderboard | |
| 10 | | 65.2 | Feb 2024 | Official leaderboard | |
| 11 | internlm2-chat-20B | 64.5 | Official leaderboard | ||
| 12 | internlm2-chat-7B | 61.8 | Official leaderboard | ||
| 13 | Llama-2-70B-chat | 60.1 | Official leaderboard | ||
| 14 | vicuna-33B-v1.3 | 58.9 | Official leaderboard | ||
| 15 | Baichuan2-7B-Chat | 57.3 | Official leaderboard | ||
| 16 | vicuna-13B-v1.5 | 54.8 | Official leaderboard | ||
| 17 | ChatGLM3-6B | 53.4 | Oct 2023 | Official leaderboard | |
| 18 | vicuna-7B-v1.5 | 47.9 | Official leaderboard | ||
| 19 | deepseek-llm-7B-chat | 45.9 | Official leaderboard | ||
| 20 | Llama-2-13B-chat | 34.5 | Official leaderboard | ||
| 21 | Llama-2-7B-chat | 28.6 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4 Turbo (Jan 2024) | 80.2 | Jan 2024 | Official leaderboard | |
| 2 | GPT-4o May 2024 | 78.8 | May 2024 | Official leaderboard | |
| 3 | GPT-4 Turbo Nov 2023 | 76.4 | Nov 2023 | Official leaderboard | |
| 4 | GPT-3.5 Turbo Jan 2024 | 67.9 | Jan 2024 | Official leaderboard | |
| 5 | | 66.3 | Nov 2023 | Official leaderboard | |
| 6 | Baichuan2-13B-Chat | 65.9 | Sep 2023 | Official leaderboard | |
| 7 | | 65.4 | Feb 2024 | Official leaderboard | |
| 8 | | 65.2 | Feb 2024 | Official leaderboard | |
| 9 | ChatGLM3-6B | 53.4 | Oct 2023 | Official leaderboard |
Rankings by part
We read 4 parts from the benchmark’s paper, each ranking models on one part of the test, such as Hallucinatory F1, Non-Hallucinatory F1 and Spontaneous Hallucination Scenario. GPT-4 Turbo (Jan 2024) leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.
Hallucinatory F1
9 models we track have a score here.
- 1 GPT-4 Turbo (Jan 2024) 74.2
- 2 GPT-4o May 2024 72.7
- 3 GPT-4 Turbo Nov 2023 67.8
- 4 Baichuan2-13B-Chat ↑2 62.1
- 5
Qwen1.5-7B-Chat ↑2 54.5
- 6 GPT-3.5 Turbo Jan 2024 ↓2 54.3
- 7
DeepSeek LLM 67B ↓2 50.8
- 8
Qwen1.5-72B 49.0
- 9 ChatGLM3-6B 34.3
Non-Hallucinatory F1
9 models we track have a score here.
- 1 GPT-4 Turbo (Jan 2024) 86.2
- 2 GPT-4 Turbo Nov 2023 ↑1 85.0
- 3 GPT-4o May 2024 ↓1 85.0
- 4
DeepSeek LLM 67B ↑1 81.9
- 5 GPT-3.5 Turbo Jan 2024 ↓1 81.4
- 6
Qwen1.5-72B ↑2 81.4
- 7
Qwen1.5-7B-Chat 76.3
- 8 ChatGLM3-6B ↑1 72.4
- 9 Baichuan2-13B-Chat ↓3 69.6
Spontaneous Hallucination Scenario
4 models we track have a score here.
- 1 GPT-4 Turbo (Jan 2024) 72.2
- 2 GPT-4o May 2024 70.7
- 3 GPT-4 Turbo Nov 2023 69.3
- 4 GPT-3.5 Turbo Jan 2024 53.5
Induced Hallucination Scenario
4 models we track have a score here.
- 1 GPT-4 Turbo (Jan 2024) 80.5
- 2 GPT-4o May 2024 78.8
- 3 GPT-4 Turbo Nov 2023 74.9
- 4 GPT-3.5 Turbo Jan 2024 67.5
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The HalluDial 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 HalluDial 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
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in May 2024, for GPT-4o. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the benchmark’s paper (21 entries).
- Pass Models scored 9 models we track, plus 12 other entries on its leaderboard.
- Warning Newest score The newest came 2 years ago, in May 2024, for GPT-4o.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-4o (#162).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 9 rows for models we track and 12 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.