Lech Mazur Creative Writing
GPT-5 8.60 leads since Aug 2025 Not in model cards
Models write short stories that must use 10 assigned elements such as characters, objects, and motivations. Judge models compare matched stories pairwise.
Earlier versions used 0 to 10 rubric ratings, which the current pairwise ranking replaces.
Made by Lech Mazur Released Jan 2025 Creative writing All tasks public
Last model added Aug 2025
Scores refreshed 31 Oct 2026
- Level
- University Undergraduates and entry-level professionals
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Open-ended Free-form text, documents, designs or media judged as a whole.
44
Models scored in Epoch AI data
GPT-5 leads with 8.60 among models released by Aug 2025
Best score per model in Epoch AI’s data. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in Aug 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 16 more
- 8.60 GPT-5
- 8.56 Kimi K2
- 8.47 Claude Opus 4.1
- 8.44 o3-pro
- 8.39 o3
- 8.38
Gemini 2.5 Pro - 8.36 Claude Opus 4
- 8.31 GPT-5 mini
- 8.30
DeepSeek-R1 - 8.30
Qwen3-235B-A22B - 8.24 Qwen3-235B-A22B-Thinking
- 8.19 DeepSeek-R1
- 8.18 GPT-4o
- 8.14 Claude Sonnet 4
- 8.11 Claude 3.7 Sonnet
- 8.09 Gemini 2.5 Pro
- 8.05 Gemini 2.5 Pro
- 8.03 Claude 3.5 Sonnet (October 2024)
- 8.02 QwQ-32B
- 7.99 Gemma 3 27B
- 7.73
Mistral Medium 3 - 7.71 gpt-oss-120b
- 7.70 DeepSeek-V3
- 7.69 Grok 4
- 7.65 Gemini 2.5 Flash
- 7.64 Grok 3
- 7.56 GPT-4.5
- 7.53 Qwen3-30B-A3B
- 7.50 o4-mini
- 7.38 Gemini 2.0 Flash Thinking
- 7.35 Claude 3.5 Haiku
- 7.35 Grok-3 mini
- 7.34 GLM-4.5
- 7.29 Qwen2.5-Max
- 7.15 Gemini 2.0 Flash
- 7.02 o1
- 6.90 Mistral Large 2
- 6.72 GPT-4o mini
- 6.49 o1-mini
- 6.36 Grok-2
- 6.26
Phi-4 - 6.20
Llama 4 Maverick - 6.17 o3-mini
- 6.05
Amazon Nova Pro
- 1 GPT-5 medium 8.60
- 2 Kimi K2 Jul 2025 8.56
- 3 Claude Opus 4.1 8.47
- 4 o3-pro medium 8.44
- 5 o3 medium 8.39
- 6 Gemini 2.5 Pro Jun 2025 8.38
- 7 Claude Opus 4 16K 8.36
- 8 GPT-5 mini medium 8.31
- 9 DeepSeek-R1 8.30
- 10 Qwen3-235B-A22B 8.30
- 11 Qwen3-235B-A22B-Thinking Jul 2025 8.24
- 12 DeepSeek-R1 May 2025 8.19
- 13 GPT-4o Nov 2024 8.18
- 14 Claude Sonnet 4 16K 8.14
- 15 Claude 3.7 Sonnet 16k thinking 8.11
- 16 Gemini 2.5 Pro May 2025 8.09
- 17 Gemini 2.5 Pro Mar 2025 8.05
- 18 Claude 3.5 Sonnet (October 2024) 8.03
- 19 QwQ-32B 8.02
- 20 Gemma 3 27B 7.99
- 21 Mistral Medium 3 7.73
- 22 gpt-oss-120b 7.71
- 23 DeepSeek-V3 Mar 2025 7.70
- 24 Grok 4 7.69
- 25 Gemini 2.5 Flash Apr 2025 7.65
- 26 Grok 3 7.64
- 27 GPT-4.5 7.56
- 28 Qwen3-30B-A3B 7.53
- 29 o4-mini medium 7.50
- 30 Gemini 2.0 Flash Thinking Jan 2025 7.38
- 31 Claude 3.5 Haiku Oct 2024 7.35
- 32 Grok-3 mini low 7.35
- 33 GLM-4.5 7.34
- 34 Qwen2.5-Max 7.29
- 35 Gemini 2.0 Flash Dec 2024 7.15
- 36 o1 medium 7.02
- 37 Mistral Large 2 Jul 2024 6.90
- 38 GPT-4o mini 6.72
- 39 o1-mini medium 6.49
- 40 Grok-2 Dec 2024 6.36
- 41 Phi-4 6.26
- 42 Llama 4 Maverick 6.20
- 43 o3-mini high 6.17
- 44 Amazon Nova Pro 6.05
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 8.60 | medium | Aug 2025 | Epoch AI |
| 2 | Kimi K2 Jul 2025 | 8.56 | Jul 2025 | Epoch AI | |
| 3 | Claude Opus 4.1 | 8.47 | Aug 2025 | Epoch AI | |
| 4 | o3-pro | 8.44 | medium | Jun 2025 | Epoch AI |
| 5 | o3 | 8.39 | medium | Apr 2025 | Epoch AI |
| 6 | Gemini 2.5 Pro Jun 2025 | 8.38 | Jun 2025 | Epoch AI | |
| 7 | Claude Opus 4 | 8.36 | 16K | May 2025 | Epoch AI |
| 8 | GPT-5 mini | 8.31 | medium | Aug 2025 | Epoch AI |
| 9 | DeepSeek-R1 | 8.30 | Jan 2025 | Epoch AI | |
| 10 | Qwen3-235B-A22B | 8.30 | Apr 2025 | Epoch AI | |
| 11 | Qwen3-235B-A22B-Thinking Jul 2025 | 8.24 | Jul 2025 | Epoch AI | |
| 12 | DeepSeek-R1 May 2025 | 8.19 | May 2025 | Epoch AI | |
| 13 | GPT-4o Nov 2024 | 8.18 | Nov 2024 | Epoch AI | |
| 14 | Claude Sonnet 4 | 8.14 | 16K | May 2025 | Epoch AI |
| 15 | Claude 3.7 Sonnet | 8.11 | 16k thinking | Feb 2025 | Epoch AI |
| 16 | Gemini 2.5 Pro May 2025 | 8.09 | May 2025 | Epoch AI | |
| 17 | Gemini 2.5 Pro Mar 2025 | 8.05 | Mar 2025 | Epoch AI | |
| 18 | Claude 3.5 Sonnet (October 2024) | 8.03 | Oct 2024 | Epoch AI | |
| 19 | QwQ-32B | 8.02 | Mar 2025 | Epoch AI | |
| 20 | Gemma 3 27B | 7.99 | Mar 2025 | Epoch AI | |
| 21 | Mistral Medium 3 | 7.73 | May 2025 | Epoch AI | |
| 22 | gpt-oss-120b | 7.71 | Aug 2025 | Epoch AI | |
| 23 | DeepSeek-V3 Mar 2025 | 7.70 | Mar 2025 | Epoch AI | |
| 24 | Grok 4 | 7.69 | Jul 2025 | Epoch AI | |
| 25 | Gemini 2.5 Flash Apr 2025 | 7.65 | Apr 2025 | Epoch AI | |
| 26 | Grok 3 | 7.64 | Apr 2025 | Epoch AI | |
| 27 | GPT-4.5 | 7.56 | Feb 2025 | Epoch AI | |
| 28 | Qwen3-30B-A3B | 7.53 | Apr 2025 | Epoch AI | |
| 29 | o4-mini | 7.50 | medium | Apr 2025 | Epoch AI |
| 30 | Gemini 2.0 Flash Thinking Jan 2025 | 7.38 | Jan 2025 | Epoch AI | |
| 31 | Claude 3.5 Haiku | 7.35 | Oct 2024 | Oct 2024 | Epoch AI |
| 32 | Grok-3 mini | 7.35 | low | Apr 2025 | Epoch AI |
| 33 | GLM-4.5 | 7.34 | Aug 2025 | Epoch AI | |
| 34 | Qwen2.5-Max | 7.29 | Jan 2025 | Epoch AI | |
| 35 | Gemini 2.0 Flash Dec 2024 | 7.15 | Dec 2024 | Epoch AI | |
| 36 | o1 | 7.02 | medium | Dec 2024 | Epoch AI |
| 37 | Mistral Large 2 Jul 2024 | 6.90 | Jul 2024 | Epoch AI | |
| 38 | GPT-4o mini | 6.72 | Jul 2024 | Epoch AI | |
| 39 | o1-mini | 6.49 | medium | Sep 2024 | Epoch AI |
| 40 | Grok-2 Dec 2024 | 6.36 | Dec 2024 | Epoch AI | |
| 41 | Phi-4 | 6.26 | Dec 2024 | Epoch AI | |
| 42 | Llama 4 Maverick | 6.20 | Apr 2025 | Epoch AI | |
| 43 | o3-mini | 6.17 | high | Jan 2025 | Epoch AI |
| 44 | Amazon Nova Pro | 6.05 | Dec 2024 | Epoch AI |
6 models have held number one since its release in Jan 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
OpenAI DeepSeek Moonshot AI
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Run or collected by Epoch AI
No lab model card reports it yet
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.
More in Writing and taste
- Design Bench (Moonshot) In use Models make graphic designs from direct prompts or few-shot reference images. An automated grader rates the results as Elo scores. Kimi K3.1 Pro 957
- WritingBench Fading Models complete 1,000 real-world writing requests across 6 domains and 100 subdomains. A judge model scores each reply on five query-specific criteria. GPT-5 83.9
- Creative Writing (MiniMax in-house) Only in system cards Models write creative pieces for real Hailuo AI user requests. Automatic assessments score the writing.
MiniMax-Text-01 81.3% - DeepSeek Chinese functional writing evaluation Only in system cards Models answer 3,170 everyday Chinese writing prompts such as reports, emails and speeches. Pairwise judging against a baseline model gives a win rate. DeepSeek-V4-Pro 62.6%
- Slide generation human evaluation (Z.ai) Only in system cards People compare HTML slide decks from two models on content, layout and visual quality. Metric is overall win rate. GLM-5 67.5%
- Task-dependent output homogenization evaluation Only in system cards Models answer prompts several times. A judge scores functional diversity where it is wanted and consistency where one answer is right. Muse Spark 0.21