Benchmark Almanac

UHGEval

GPT-4 Jun 2023 0.70 leads since November 2023
Not in model cards

Models identify hallucinated Chinese text or continue a supplied passage. Classification uses accuracy, while generation uses BLEU-4, ROUGE-L, keyword precision, and BERTScore.

The full set has 5,141 items. A 1,000-item subset reduces evaluation cost.

Released
November 2023
Kind
Hallucination rates
Level
Everyday
Scored
Checked automatically
Answers
Open-ended
Tasks
All public
Measures
Propensity
Behavior
Hallucination
Language
Chinese
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GPT-4 Turbo Nov 2023 #176 on the capability index
Last model added
June 2024
Scores refreshed
November 20th, 2026

In the authors’ words

Large language models (LLMs) have emerged as pivotal contributors in contemporary natural language processing and are increasingly being applied across a diverse range of industries. However, these large-scale probabilistic statistical models cannot currently ensure the requisite quality in professional content generation. These models often produce hallucinated text, compromising their practical utility in professional contexts. To assess the authentic reliability of LLMs in text generation, numerous initiatives have developed benchmark evaluations for hallucination phenomena. Nevertheless, these benchmarks frequently utilize constrained generation techniques due to cost and temporal constraints. These techniques encompass the use of directed hallucination induction and strategies that deliberately alter authentic text to produce hallucinations. These approaches are not congruent with the unrestricted text generation demanded by real-world applications. Furthermore, a well-established Chinese-language dataset dedicated to the evaluation of hallucinations in text generation is presently lacking. Consequently, we have developed an Unconstrained Hallucination Generation Evaluation (UHGEval) benchmark, designed to compile outputs produced with minimal restrictions by LLMs. Concurrently, we have established a comprehensive benchmark evaluation framework to aid subsequent researchers in undertaking scalable and reproducible experiments. We have also executed extensive experiments, evaluating prominent Chinese language models and the GPT series models to derive professional performance insights regarding hallucination challenges.

Abstract of UHGEval: Benchmarking the Hallucination of Chinese Large Language Models via Unconstrained Generation (arXiv:2311.15296, CC0)
14
Entries in its paper, 6 models we track

GPT-4 leads with 0.70 among models released by June 2024 GPT-4 leads with 0.70 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 discriminative-keyword average accuracy. 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 June 2024, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-4 Jun 2023, November 2023 full-dataset evaluation 0.70
  2. 2 GPT-4 Turbo Nov 2023, November 2023 full-dataset evaluation 0.69
  3. 3 gpt-3.5-turbo (snapshot unspecified) November 2023 full-dataset evaluation 0.54
  4. 4 Aquila_34B_Chat November 2023 full-dataset evaluation 0.54
  5. 5 Xinyu_70B_Chat November 2023 full-dataset evaluation 0.53
  6. 6 Qwen-14B November 2023 full-dataset evaluation 0.53
  7. 7 Qwen1.5-7B-Chat August 2024 full-dataset evaluation 0.53
  8. 8 Baichuan2_53B_Chat November 2023 full-dataset evaluation 0.52
  9. 9 Baichuan2-13B-Chat November 2023 full-dataset evaluation 0.52
  10. 10 GLM-4-9B-Chat August 2024 full-dataset evaluation 0.52
  11. 11 InternLM_20B_Chat November 2023 full-dataset evaluation 0.51
  12. 12 ChatGLM2_6B_Chat November 2023 full-dataset evaluation 0.51
  13. 13 Xinyu_7B_Chat November 2023 full-dataset evaluation 0.50
  14. 14 meta-llama/Meta-Llama-3.1-8B-Instruct August 2024 full-dataset evaluation 0.49

Source: The UHGEval paper

#EntryScoreSettingDateSource
1 GPT-4 Jun 2023 0.70 November 2023 full-dataset evaluation Jun 2023 Official leaderboard
2 GPT-4 Turbo Nov 2023 0.69 November 2023 full-dataset evaluation Nov 2023 Official leaderboard
3 gpt-3.5-turbo (snapshot unspecified) 0.54 November 2023 full-dataset evaluation Official leaderboard
4 Aquila_34B_Chat 0.54 November 2023 full-dataset evaluation Official leaderboard
5 Xinyu_70B_Chat 0.53 November 2023 full-dataset evaluation Official leaderboard
6 Qwen-14B 0.53 November 2023 full-dataset evaluation Sep 2023 Official leaderboard
7 Qwen1.5-7B-Chat 0.53 August 2024 full-dataset evaluation Feb 2024 Official leaderboard
8 Baichuan2_53B_Chat 0.52 November 2023 full-dataset evaluation Official leaderboard
9 Baichuan2-13B-Chat 0.52 November 2023 full-dataset evaluation Sep 2023 Official leaderboard
10 GLM-4-9B-Chat 0.52 August 2024 full-dataset evaluation Jun 2024 Official leaderboard
11 InternLM_20B_Chat 0.51 November 2023 full-dataset evaluation Official leaderboard
12 ChatGLM2_6B_Chat 0.51 November 2023 full-dataset evaluation Official leaderboard
13 Xinyu_7B_Chat 0.50 November 2023 full-dataset evaluation Official leaderboard
14 meta-llama/Meta-Llama-3.1-8B-Instruct 0.49 August 2024 full-dataset evaluation Official leaderboard
#ModelScoreSettingDateSource
1 GPT-4 Jun 2023 0.70 November 2023 full-dataset evaluation Jun 2023 Official leaderboard
2 GPT-4 Turbo Nov 2023 0.69 November 2023 full-dataset evaluation Nov 2023 Official leaderboard
3 Qwen-14B 0.53 November 2023 full-dataset evaluation Sep 2023 Official leaderboard
4 Qwen1.5-7B-Chat 0.53 August 2024 full-dataset evaluation Feb 2024 Official leaderboard
5 Baichuan2-13B-Chat 0.52 November 2023 full-dataset evaluation Sep 2023 Official leaderboard
6 GLM-4-9B-Chat 0.52 August 2024 full-dataset evaluation Jun 2024 Official leaderboard

Rankings by part

We read 7 parts from the benchmark’s paper, each ranking models on one part of the test, such as Keyword-level discrimination accuracy, Sentence-level discrimination accuracy and Selection accuracy. GPT-4 leads 2 of them, and 4 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Keyword-level discrimination accuracy

6 models we track have a score here.

  1. 1 GPT-4 Jun 2023 0.70
  2. 2 GPT-4 Turbo Nov 2023 0.69
  3. 3 Qwen-14B 0.53
  4. 4 Qwen1.5-7B-Chat 0.53
  5. 5 Baichuan2-13B-Chat 0.52
  6. 6 GLM-4-9B-Chat 0.52

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Benchmark paper Not tested since Jun 2024
GPT-4 0.70

Source: The UHGEval paper

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper
0.55 0.60 0.65 0.70 0.75 2023 2024 0.55 0.60 0.65 0.70 0.75 2023 2024

Source: Official UHGEval 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

All 164 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 June 2024, for GLM-4-9B-Chat. 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 (14 entries).
  • Pass Models scored 6 models we track, plus 8 other entries on its leaderboard.
  • Warning Newest score The newest came 2 years ago, in June 2024, for GLM-4-9B-Chat.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-4 Turbo (#176).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on github.com: 6 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.

How we check