Benchmark Almanac

NoveltyBench

novamine-ai + Claude Opus 5 6.65 tops its leaderboard Claude Fable 5.1 5.32 leads as a model on its own
Not in model cards

Models give several answers to the same open-ended request. The benchmark measures how many distinct responses they produce and how useful those responses are.

Made by
Yiming Zhang and colleagues
Released
April 2025
Kind
Creative writing
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
September 2026
Scores refreshed
November 20th, 2026
33
Entries on its own leaderboard, 27 models we track

novamine-ai + Claude Opus 5 tops the leaderboard with 6.65 Claude Fable 5.1 leads with 5.32 among models released by September 2026

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran on its own with several settings, we keep its best. Every run is in the next chart. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. The best model on its own is Claude Fable 5.1, at 5.32. Agents built around a model, such as novamine-ai + Claude Opus 5, are their own entries. We don't count them as the model's score. This view shows only models we track, each at its best.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 novamine-ai + Claude Opus 5 regenerate, judge v1.1 6.65
  2. 2 novamine-ai-trill + Claude Fable 5 regenerate, judge v1.1 6.25
  3. 3 Claude Fable 5.1 in-context, judge v1.1 5.32
  4. 4 Claude Opus 5 in-context, judge v1.1 4.94
  5. 5 GPT-6 Astra in-context, judge v1.1 4.45
  6. 6 GPT-5.6 Sol in-context, judge v1.1 4.28
  7. 7 GPT-5.6 Terra in-context, judge v1.1 4.05
  8. 8 GPT-5.6 Luna in-context, judge v1.1 3.70
  9. 9 CrPO-llama-3.1-8b-instruct regenerate, judge v1.1 2.18
  10. 10 Gemma 2 27B regenerate, judge v1.1 2.00
  11. 11 Gemma 2 9B regenerate, judge v1.1 1.99
  12. 12 gemini-1.5-pro (snapshot unspecified) regenerate, judge v1.1 1.94
  13. 13 Llama 3.1-405B regenerate, judge v1.1 1.87
  14. 14 Claude 3 Opus regenerate, judge v1.1 1.84
  15. 15 Claude 3.5 Sonnet (October 2024) regenerate, judge v1.1 1.82
  16. 16 GPT-4o Nov 2024, regenerate, judge v1.1 1.82
  17. 17 GPT-4o mini regenerate, judge v1.1 1.81
  18. 18 Llama 3.1-8B regenerate, judge v1.1 1.81
  19. 19 Gemini 2.0 Flash Feb 2025, regenerate, judge v1.1 1.78
  20. 20 Gemma 2 2B IT regenerate, judge v1.1 1.76
  21. 21 Gemini 2.0 Flash-Lite regenerate, judge v1.1 1.76
  22. 22 c4ai-command-r-08-2024 regenerate, judge v1.1 1.71
  23. 23 Command R+ regenerate, judge v1.1 1.71
  24. 24 Llama 3.3 70B Instruct regenerate, judge v1.1 1.69
  25. 25 Claude 3.5 Haiku regenerate, judge v1.1 1.68
  26. 26 Llama-3.2-3B-Instruct regenerate, judge v1.1 1.65
  27. 27 Command R7B regenerate, judge v1.1 1.63
  28. 28 Gemini 2.0 Pro regenerate, judge v1.1 1.62
  29. 29 Gemma 3 4B regenerate, judge v1.1 1.55
  30. 30 Gemma 3 27B regenerate, judge v1.1 1.50
  31. 31 Llama-3.2-1B-Instruct regenerate, judge v1.1 1.48
  32. 32 Gemma 3 12B regenerate, judge v1.1 1.46
  33. 33 Gemma 3 1B regenerate, judge v1.1 1.23

Source: Official NoveltyBench leaderboard

Source: Official NoveltyBench leaderboard

#EntryScoreSettingSubmitted byReleasedRun onSource
1 novamine-ai + Claude Opus 5 6.65 regenerate, judge v1.1 Novamine-AI Jul 2026 Aug 2026 Official leaderboard
2 novamine-ai-trill + Claude Fable 5 6.25 regenerate, judge v1.1 Novamine-AI Jun 2026 Aug 2026 Official leaderboard
3 Claude Fable 5.1 5.32 in-context, judge v1.1 Sep 2026 Sep 2026 Official leaderboard
4 Claude Opus 5 4.94 in-context, judge v1.1 Jul 2026 Sep 2026 Official leaderboard
5 GPT-6 Astra 4.45 in-context, judge v1.1 Sep 2026 Sep 2026 Official leaderboard
6 GPT-5.6 Sol 4.28 in-context, judge v1.1 Jul 2026 Sep 2026 Official leaderboard
7 GPT-5.6 Terra 4.05 in-context, judge v1.1 Jul 2026 Sep 2026 Official leaderboard
8 GPT-5.6 Luna 3.70 in-context, judge v1.1 Jul 2026 Sep 2026 Official leaderboard
9 CrPO-llama-3.1-8b-instruct open weights 2.18 regenerate, judge v1.1 mismayil Nov 2025 Official leaderboard
10 Gemma 2 27B 2.00 regenerate, judge v1.1 Jun 2024 Mar 2025 Official leaderboard
11 Gemma 2 9B 1.99 regenerate, judge v1.1 Jun 2024 Mar 2025 Official leaderboard
12 gemini-1.5-pro (snapshot unspecified) closed weights 1.94 regenerate, judge v1.1 Mar 2025 Official leaderboard
13 Llama 3.1-405B 1.87 regenerate, judge v1.1 Jul 2024 Mar 2025 Official leaderboard
14 Claude 3 Opus 1.84 regenerate, judge v1.1 Feb 2024 Mar 2025 Official leaderboard
15 Claude 3.5 Sonnet (October 2024) 1.82 regenerate, judge v1.1 Oct 2024 Mar 2025 Official leaderboard
16 GPT-4o Nov 2024 1.82 regenerate, judge v1.1 Nov 2024 Mar 2025 Official leaderboard
17 GPT-4o mini 1.81 regenerate, judge v1.1 Jul 2024 Mar 2025 Official leaderboard
18 Llama 3.1-8B 1.81 regenerate, judge v1.1 Jul 2024 Mar 2025 Official leaderboard
19 Gemini 2.0 Flash Feb 2025 1.78 regenerate, judge v1.1 Feb 2025 Mar 2025 Official leaderboard
20 Gemma 2 2B IT 1.76 regenerate, judge v1.1 Jul 2024 Mar 2025 Official leaderboard
21 Gemini 2.0 Flash-Lite 1.76 regenerate, judge v1.1 Feb 2025 Mar 2025 Official leaderboard
22 c4ai-command-r-08-2024 1.71 regenerate, judge v1.1 Aug 2024 Mar 2025 Official leaderboard
23 Command R+ 1.71 regenerate, judge v1.1 Aug 2024 Mar 2025 Official leaderboard
24 Llama 3.3 70B Instruct 1.69 regenerate, judge v1.1 Dec 2024 Mar 2025 Official leaderboard
25 Claude 3.5 Haiku 1.68 regenerate, judge v1.1 Oct 2024 Mar 2025 Official leaderboard
26 Llama-3.2-3B-Instruct open weights 1.65 regenerate, judge v1.1 Mar 2025 Official leaderboard
27 Command R7B 1.63 regenerate, judge v1.1 Dec 2024 Mar 2025 Official leaderboard
28 Gemini 2.0 Pro 1.62 regenerate, judge v1.1 Feb 2025 Mar 2025 Official leaderboard
29 Gemma 3 4B 1.55 regenerate, judge v1.1 Mar 2025 Sep 2025 Official leaderboard
30 Gemma 3 27B 1.50 regenerate, judge v1.1 Mar 2025 Sep 2025 Official leaderboard
31 Llama-3.2-1B-Instruct open weights 1.48 regenerate, judge v1.1 Mar 2025 Official leaderboard
32 Gemma 3 12B 1.46 regenerate, judge v1.1 Mar 2025 Sep 2025 Official leaderboard
33 Gemma 3 1B 1.23 regenerate, judge v1.1 Mar 2025 Sep 2025 Official leaderboard
#ModelScoreSettingSubmitted byReleasedRun onSource
1 Claude Fable 5.1 5.32 in-context, judge v1.1 Sep 2026 Sep 2026 Official leaderboard
2 Claude Opus 5 4.94 in-context, judge v1.1 Jul 2026 Sep 2026 Official leaderboard
3 GPT-6 Astra 4.45 in-context, judge v1.1 Sep 2026 Sep 2026 Official leaderboard
4 GPT-5.6 Sol 4.28 in-context, judge v1.1 Jul 2026 Sep 2026 Official leaderboard
5 GPT-5.6 Terra 4.05 in-context, judge v1.1 Jul 2026 Sep 2026 Official leaderboard
6 GPT-5.6 Luna 3.70 in-context, judge v1.1 Jul 2026 Sep 2026 Official leaderboard
7 Gemma 2 27B 2.00 regenerate, judge v1.1 Jun 2024 Mar 2025 Official leaderboard
8 Gemma 2 9B 1.99 regenerate, judge v1.1 Jun 2024 Mar 2025 Official leaderboard
9 Llama 3.1-405B 1.87 regenerate, judge v1.1 Jul 2024 Mar 2025 Official leaderboard
10 Claude 3 Opus 1.84 regenerate, judge v1.1 Feb 2024 Mar 2025 Official leaderboard
11 Claude 3.5 Sonnet (October 2024) 1.82 regenerate, judge v1.1 Oct 2024 Mar 2025 Official leaderboard
12 GPT-4o Nov 2024 1.82 regenerate, judge v1.1 Nov 2024 Mar 2025 Official leaderboard
13 GPT-4o mini 1.81 regenerate, judge v1.1 Jul 2024 Mar 2025 Official leaderboard
14 Llama 3.1-8B 1.81 regenerate, judge v1.1 Jul 2024 Mar 2025 Official leaderboard
15 Gemini 2.0 Flash Feb 2025 1.78 regenerate, judge v1.1 Feb 2025 Mar 2025 Official leaderboard
16 Gemma 2 2B IT 1.76 regenerate, judge v1.1 Jul 2024 Mar 2025 Official leaderboard
17 Gemini 2.0 Flash-Lite 1.76 regenerate, judge v1.1 Feb 2025 Mar 2025 Official leaderboard
18 c4ai-command-r-08-2024 1.71 regenerate, judge v1.1 Aug 2024 Mar 2025 Official leaderboard
19 Command R+ 1.71 regenerate, judge v1.1 Aug 2024 Mar 2025 Official leaderboard
20 Llama 3.3 70B Instruct 1.69 regenerate, judge v1.1 Dec 2024 Mar 2025 Official leaderboard
21 Claude 3.5 Haiku 1.68 regenerate, judge v1.1 Oct 2024 Mar 2025 Official leaderboard
22 Command R7B 1.63 regenerate, judge v1.1 Dec 2024 Mar 2025 Official leaderboard
23 Gemini 2.0 Pro 1.62 regenerate, judge v1.1 Feb 2025 Mar 2025 Official leaderboard
24 Gemma 3 4B 1.55 regenerate, judge v1.1 Mar 2025 Sep 2025 Official leaderboard
25 Gemma 3 27B 1.50 regenerate, judge v1.1 Mar 2025 Sep 2025 Official leaderboard
26 Gemma 3 12B 1.46 regenerate, judge v1.1 Mar 2025 Sep 2025 Official leaderboard
27 Gemma 3 1B 1.23 regenerate, judge v1.1 Mar 2025 Sep 2025 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude Fable 5.1
regenerate,… in-context,…
5.32
Claude Opus 5
regenerate,… in-context,…
4.94
GPT-6 Astra
regenerate,… in-context,…
4.45
GPT-5.6 Sol
regenerate,… in-context,…
4.28
GPT-5.6 Terra
regenerate,… in-context,…
4.05
GPT-5.6 Luna
regenerate,… in-context,…
3.70

Source: Official NoveltyBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
novamine-ai + Claude Opus 5 regenerate, judge v1.1 novamine-ai 6.65
novamine-ai-trill + Claude Fable 5 regenerate, judge v1.1 novamine-ai-trill 6.25
Claude Fable 5.1 in-context, judge v1.1 5.32
Claude Opus 5 in-context, judge v1.1 4.94
GPT-6 Astra in-context, judge v1.1 4.45
GPT-5.6 Sol in-context, judge v1.1 4.28
GPT-5.6 Terra in-context, judge v1.1 4.05
GPT-5.6 Luna in-context, judge v1.1 3.70
Claude Opus 5 regenerate, judge v1.1 2.90
Claude Fable 5.1 regenerate, judge v1.1 2.85
GPT-6 Astra regenerate, judge v1.1 2.67
GPT-5.6 Sol regenerate, judge v1.1 2.62
GPT-5.6 Terra regenerate, judge v1.1 2.62
GPT-5.6 Luna regenerate, judge v1.1 2.53
Gemma 2 27B regenerate, judge v1.1 2.00
Gemma 2 9B regenerate, judge v1.1 1.99
Llama 3.1-405B regenerate, judge v1.1 1.87
Claude 3 Opus regenerate, judge v1.1 1.84
Claude 3.5 Sonnet (October 2024) regenerate, judge v1.1 1.82
GPT-4o Nov 2024 regenerate, judge v1.1 1.82
GPT-4o mini regenerate, judge v1.1 1.81
Llama 3.1-8B regenerate, judge v1.1 1.81
Gemini 2.0 Flash Feb 2025 regenerate, judge v1.1 1.78
Gemma 2 2B IT regenerate, judge v1.1 1.76
Gemini 2.0 Flash-Lite regenerate, judge v1.1 1.76
c4ai-command-r-08-2024 regenerate, judge v1.1 1.71
Command R+ regenerate, judge v1.1 1.71
Llama 3.3 70B Instruct regenerate, judge v1.1 1.69
Claude 3.5 Haiku regenerate, judge v1.1 1.68
Command R7B regenerate, judge v1.1 1.63
Gemini 2.0 Pro regenerate, judge v1.1 1.62
Gemma 3 4B regenerate, judge v1.1 1.55
Gemma 3 27B regenerate, judge v1.1 1.50
Gemma 3 12B regenerate, judge v1.1 1.46
Gemma 3 1B regenerate, judge v1.1 1.23

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude Fable 5.1
regenerate,… in-context,…
5.32
Claude Opus 5
regenerate,… in-context,…
4.94
GPT-6 Astra
regenerate,… in-context,…
4.45
GPT-5.6 Sol
regenerate,… in-context,…
4.28
GPT-5.6 Terra
regenerate,… in-context,…
4.05
GPT-5.6 Luna
regenerate,… in-context,…
3.70

Source: Official NoveltyBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5.1 in-context, judge v1.1 5.32
Claude Opus 5 in-context, judge v1.1 4.94
GPT-6 Astra in-context, judge v1.1 4.45
GPT-5.6 Sol in-context, judge v1.1 4.28
GPT-5.6 Terra in-context, judge v1.1 4.05
GPT-5.6 Luna in-context, judge v1.1 3.70
Claude Opus 5 regenerate, judge v1.1 2.90
Claude Fable 5.1 regenerate, judge v1.1 2.85
GPT-6 Astra regenerate, judge v1.1 2.67
GPT-5.6 Sol regenerate, judge v1.1 2.62
GPT-5.6 Terra regenerate, judge v1.1 2.62
GPT-5.6 Luna regenerate, judge v1.1 2.53
Gemma 2 27B regenerate, judge v1.1 2.00
Gemma 2 9B regenerate, judge v1.1 1.99
Llama 3.1-405B regenerate, judge v1.1 1.87
Claude 3 Opus regenerate, judge v1.1 1.84
Claude 3.5 Sonnet (October 2024) regenerate, judge v1.1 1.82
GPT-4o Nov 2024 regenerate, judge v1.1 1.82
GPT-4o mini regenerate, judge v1.1 1.81
Llama 3.1-8B regenerate, judge v1.1 1.81
Gemini 2.0 Flash Feb 2025 regenerate, judge v1.1 1.78
Gemma 2 2B IT regenerate, judge v1.1 1.76
Gemini 2.0 Flash-Lite regenerate, judge v1.1 1.76
c4ai-command-r-08-2024 regenerate, judge v1.1 1.71
Command R+ regenerate, judge v1.1 1.71
Llama 3.3 70B Instruct regenerate, judge v1.1 1.69
Claude 3.5 Haiku regenerate, judge v1.1 1.68
Command R7B regenerate, judge v1.1 1.63
Gemini 2.0 Pro regenerate, judge v1.1 1.62
Gemma 3 4B regenerate, judge v1.1 1.55
Gemma 3 27B regenerate, judge v1.1 1.50
Gemma 3 12B regenerate, judge v1.1 1.46
Gemma 3 1B regenerate, judge v1.1 1.23

4 models have held number one on the official leaderboard since its release in April 2025

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

Google DeepMind OpenAI Anthropic
Own leaderboard Not tested since Sep 2026

Source: Official NoveltyBench leaderboard

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
On the benchmark’s own leaderboard
1.00 2.00 3.00 4.00 5.00 6.00 2024 2025 2026 1.00 2.00 3.00 4.00 5.00 6.00 2024 2025 2026

Source: Official NoveltyBench 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 Writing and taste

All 65 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

Data audit

All 9 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (33 entries).
  • Pass Models scored 27 models we track, plus 6 other entries on its leaderboard.
  • Pass Newest score The newest is from September 2026, for GPT-6 Astra.
  • Note Frontier models Only 2 of the top 10 have a score. The strongest tested is GPT-6 Astra (#4).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on novelty-bench.github.io: 35 rows for models we track and 4 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