HelloBench
Models generate long answers, summaries, conversations, text completions, and texts from writing instructions. Model judges apply task-specific checklists to the responses.
Generation limits differ across models. The released prompts include separate length-constrained settings.
- Made by
- Beihang University, Shanghai AI Laboratory, Peking University and others
- Released
- September 2024
- Kind
- Overall chat quality
- Level
- University
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Subject
- Languages and literature
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- o1-mini #142 on the capability index
- Last model added
- September 2024
- Scores refreshed
- November 18th, 2026
In the authors’ words
In recent years, Large Language Models (LLMs) have demonstrated remarkable capabilities in various tasks (e.g., long-context understanding), and many benchmarks have been proposed. However, we observe that long text generation capabilities are not well investigated. Therefore, we introduce the Hierarchical Long Text Generation Benchmark (HelloBench), a comprehensive, in-the-wild, and open-ended benchmark to evaluate LLMs' performance in generating long text. Based on Bloom's Taxonomy, HelloBench categorizes long text generation tasks into five subtasks: open-ended QA, summarization, chat, text completion, and heuristic text generation. Besides, we propose Hierarchical Long Text Evaluation (HelloEval), a human-aligned evaluation method that significantly reduces the time and effort required for human evaluation while maintaining a high correlation with human evaluation. We have conducted extensive experiments across around 30 mainstream LLMs and observed that the current LLMs lack long text generation capabilities. Specifically, first, regardless of whether the instructions include explicit or implicit length constraints, we observe that most LLMs cannot generate text that is longer than 4000 words. Second, we observe that while some LLMs can generate longer text, many issues exist (e.g., severe repetition and quality degradation). Third, to demonstrate the effectiveness of HelloEval, we compare HelloEval with traditional metrics (e.g., ROUGE, BLEU, etc.) and LLM-as-a-Judge methods, which show that HelloEval has the highest correlation with human evaluation. We release our code in https://github.com/Quehry/HelloBench.
GPT-4o leads with 48.5 among models released by September 2024 GPT-4o leads with 48.5 among models released by September 2024
Best score per entry from the benchmark's paper, read on November 18th, 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. This view shows only models we track, each at its best.
The newest model with a score here came out in September 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
- 48.5 GPT-4o
- 46.8 Mistral-Large-API
- 46.1 o1-mini
- 43.8 Claude-3.5-Sonnet (snapshot unspecified)
- 39.2 Gemini-1.5-Pro (snapshot unspecified)
- 35.7 Deepseek-API
- 35.3
Gemma 2 27B
- 33.0
Yi-Large
- 30.6
Llama 3.1-70B
- 30.2
Qwen2-72B
- 28.4 Qwen-Max (snapshot unspecified)
- 28.1 InternLM-2.5-20B
- 24.8 GLM-4-API
- 23.6
Yi-1.5-34B
- 14.5
Llama 3.1-8B
- 13.8 GLM-4-9B
- 13.7
Qwen2-7B
- 12.9 InternLM-2.5-7B
- 10.0 Mistral-7B-0.2
- 8.33 LongWriter-GLM4-9B
- 8.16
Phi-3.5-MoE
- 0.03 MAP-Neo
- -83.58 Suri-I-ORPO
- 1 GPT-4o Aug 2024 48.5
- 2 Mistral-Large-API 46.8
- 3 o1-mini 46.1
- 4 Claude-3.5-Sonnet (snapshot unspecified) 43.8
- 5 Gemini-1.5-Pro (snapshot unspecified) 39.2
- 6 Deepseek-API 35.7
- 7
Gemma 2 27B 35.3
- 8
Yi-Large 33.0
- 9
Llama 3.1-70B 30.6
- 10
Qwen2-72B 30.2
- 11 Qwen-Max (snapshot unspecified) 28.4
- 12 InternLM-2.5-20B 28.1
- 13 GLM-4-API 24.8
- 14
Yi-1.5-34B 23.6
- 15
Llama 3.1-8B 14.5
- 16 GLM-4-9B 13.8
- 17
Qwen2-7B 13.7
- 18 InternLM-2.5-7B 12.9
- 19 Mistral-7B-0.2 10.0
- 20 LongWriter-GLM4-9B 8.33
- 21
Phi-3.5-MoE 8.16
- 22 MAP-Neo 0.03
- 23 Suri-I-ORPO -83.58
Source: The HelloBench paper
- 1 GPT-4o Aug 2024 48.5
- 2 o1-mini 46.1
- 3
Gemma 2 27B 35.3
- 4
Yi-Large 33.0
- 5
Llama 3.1-70B 30.6
- 6
Qwen2-72B 30.2
- 7
Yi-1.5-34B 23.6
- 8
Llama 3.1-8B 14.5
- 9 GLM-4-9B 13.8
- 10
Qwen2-7B 13.7
- 11
Phi-3.5-MoE 8.16
Source: The HelloBench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4o Aug 2024 | 48.5 | Aug 2024 | Official leaderboard | |
| 2 | Mistral-Large-API | 46.8 | Official leaderboard | ||
| 3 | o1-mini | 46.1 | Sep 2024 | Official leaderboard | |
| 4 | Claude-3.5-Sonnet (snapshot unspecified) | 43.8 | Official leaderboard | ||
| 5 | Gemini-1.5-Pro (snapshot unspecified) | 39.2 | Official leaderboard | ||
| 6 | Deepseek-API | 35.7 | Official leaderboard | ||
| 7 | | 35.3 | Jun 2024 | Official leaderboard | |
| 8 | | 33.0 | May 2024 | Official leaderboard | |
| 9 | | 30.6 | Jul 2024 | Official leaderboard | |
| 10 | | 30.2 | Jun 2024 | Official leaderboard | |
| 11 | Qwen-Max (snapshot unspecified) | 28.4 | Official leaderboard | ||
| 12 | InternLM-2.5-20B | 28.1 | Official leaderboard | ||
| 13 | GLM-4-API | 24.8 | Official leaderboard | ||
| 14 | | 23.6 | May 2024 | Official leaderboard | |
| 15 | | 14.5 | Jul 2024 | Official leaderboard | |
| 16 | GLM-4-9B | 13.8 | Jun 2024 | Official leaderboard | |
| 17 | | 13.7 | Jun 2024 | Official leaderboard | |
| 18 | InternLM-2.5-7B | 12.9 | Official leaderboard | ||
| 19 | Mistral-7B-0.2 | 10.0 | Official leaderboard | ||
| 20 | LongWriter-GLM4-9B | 8.33 | Official leaderboard | ||
| 21 | | 8.16 | Aug 2024 | Official leaderboard | |
| 22 | MAP-Neo | 0.03 | Official leaderboard | ||
| 23 | Suri-I-ORPO | -83.58 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4o Aug 2024 | 48.5 | Aug 2024 | Official leaderboard | |
| 2 | o1-mini | 46.1 | Sep 2024 | Official leaderboard | |
| 3 | | 35.3 | Jun 2024 | Official leaderboard | |
| 4 | | 33.0 | May 2024 | Official leaderboard | |
| 5 | | 30.6 | Jul 2024 | Official leaderboard | |
| 6 | | 30.2 | Jun 2024 | Official leaderboard | |
| 7 | | 23.6 | May 2024 | Official leaderboard | |
| 8 | | 14.5 | Jul 2024 | Official leaderboard | |
| 9 | GLM-4-9B | 13.8 | Jun 2024 | Official leaderboard | |
| 10 | | 13.7 | Jun 2024 | Official leaderboard | |
| 11 | | 8.16 | Aug 2024 | Official leaderboard |
Rankings by part
We read 5 parts from the benchmark’s paper, each ranking models on one part of the test, such as Question answering, Summarization and Chat. GPT-4o leads 3 of them, and o1-mini leads the other 2. Arrows show how far a model moved from its overall rank among the models we track.
Question answering
11 models we track have a score here.
- 1 GPT-4o Aug 2024 54.8
- 2
Phi-3.5-MoE ↑9 54.3
- 3
Gemma 2 27B 52.4
- 4
Qwen2-72B ↑2 48.8
- 5
Yi-Large ↓1 48.3
- 6
Llama 3.1-70B ↓1 48.1
- 7
Yi-1.5-34B 47.4
- 8 o1-mini ↓6 46.9
- 9
Qwen2-7B ↑1 46.0
- 10
Llama 3.1-8B ↓2 42.5
Summarization
11 models we track have a score here.
- 1 o1-mini ↑1 38.6
- 2 GPT-4o Aug 2024 ↓1 29.7
- 3
Qwen2-72B ↑3 26.6
- 4
Yi-Large 23.1
- 5
Llama 3.1-70B 20.7
- 6
Gemma 2 27B ↓3 17.8
- 7
Llama 3.1-8B ↑1 15.8
- 8
Qwen2-7B ↑2 7.37
- 9
Phi-3.5-MoE ↑2 -3.70
- 10 GLM-4-9B ↓1 -5.38
Chat
11 models we track have a score here.
- 1 GPT-4o Aug 2024 42.9
- 2 o1-mini 38.8
- 3
Llama 3.1-70B ↑2 27.0
- 4
Gemma 2 27B ↓1 18.1
- 5
Yi-Large ↓1 16.5
- 6
Qwen2-72B 5.04
- 7
Yi-1.5-34B 5.02
- 8 GLM-4-9B ↑1 0.47
- 9
Llama 3.1-8B ↓1 -5.26
- 10
Qwen2-7B -6.48
Completion
11 models we track have a score here.
- 1 GPT-4o Aug 2024 67.5
- 2 o1-mini 57.5
- 3
Yi-Large ↑1 45.8
- 4
Yi-1.5-34B ↑3 44.7
- 5
Gemma 2 27B ↓2 41.8
- 6
Qwen2-72B 34.9
- 7
Llama 3.1-70B ↓2 25.3
- 8 GLM-4-9B ↑1 12.3
- 9
Qwen2-7B ↑1 5.12
- 10
Llama 3.1-8B ↓2 -5.61
Heuristic generation
11 models we track have a score here.
- 1 o1-mini ↑1 48.8
- 2 GPT-4o Aug 2024 ↓1 47.9
- 3
Gemma 2 27B 46.3
- 4
Qwen2-72B ↑2 35.7
- 5
Yi-1.5-34B ↑2 35.3
- 6
Llama 3.1-70B ↓1 31.8
- 7
Yi-Large ↓3 31.2
- 8
Llama 3.1-8B 25.0
- 9 GLM-4-9B 21.1
- 10
Qwen2-7B 16.3
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The HelloBench 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 HelloBench 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 Instructions and chat
- IFEval In use Models answer about 500 prompts containing verifiable instructions such as word limits or required keywords. Code checks each instruction for strict and loose accuracy.
Gemma 4 31B IT 98.9%
- IFBench In use 58 new out-of-domain verifiable output constraints. Scored by programmatic constraint checks (strict or loose). Flint 89.2%
- MultiChallenge In use Multi-turn conversations testing instruction retention, inference memory, versioned editing and self-coherence. Graded by LLM judge with rubrics. GPT-6.1 Sol 82%
- Multi-IF In use Models follow verifiable instructions over three-turn conversations in eight languages, extending IFEval. Scored by instruction accuracy at each turn. LFM3-1.7B 89.1%
- AdvancedIF In use Models respond to over 1,600 prompts with complex, multi-turn, and system-level instructions. A judge model checks each response against expert-written rubrics.
MAI-Thinking-1 85%
- MM-IFEval In use Multimodal models follow image-grounded instructions with exact output constraints. Checks mix rule-based verification and a judge model. GPT-4o 64.6%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in September 2024, for o1-mini. 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 (23 entries).
- Pass Models scored 11 models we track, plus 12 other entries on its leaderboard.
- Warning Newest score The newest came 2 years ago, in September 2024, for o1-mini.
- Note Frontier models None of the top 10 has a score. The strongest tested is o1-mini (#142).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 11 rows for models we track and 12 other entries at the last read.
- Pass Last read November 18th, 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.