Fiction Bench (Foreverse)
Models continue two Chinese novels for 20 rounds and receive their earlier writing as context. Reviewers rank anonymous continuations for each genre.
The maker ranks nine models by genre and preserves disputed rank ranges. The leaderboard identifies agent judges. The repository methodology identifies human judges.
- Made by
- Foreverse Research
- Released
- July 2026
- Kind
- Creative writing
- Level
- University
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Preference
- Language
- Chinese
- Interaction
- Multi-turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- GPT-5.6 Terra #18 on the capability index
- Last model added
- July 2026
- Scores refreshed
- November 20th, 2026
3 models share the top score of 2.50 among models released by July 2026
Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Plain mean of four blind review ranks across two genres, lower is better. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in July 2026, 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 15 more
- 1
DeepSeek-V4-Pro 2.50
- 2 GLM-5.2 2.50
- 3 GPT-5.6 Terra 2.50
- 4
DeepSeek-V4-Flash 3.75
- 5 Claude Opus 4.8 5.00
- 6 Kimi K2.6 6.50
- 7
Qwen3.7-Max 6.75
- 8
Gemini 3.1 Pro 7.00
- 9 Grok 4.5 8.50
Source: Official Fiction Bench (Foreverse) leaderboard
| # | Model | Score | Setting | Released | Run on | Source |
|---|---|---|---|---|---|---|
| 1 | | 2.50 | Apr 2026 | Jul 2026 | Official leaderboard | |
| 2 | GLM-5.2 | 2.50 | Jun 2026 | Jul 2026 | Official leaderboard | |
| 3 | GPT-5.6 Terra | 2.50 | Jul 2026 | Jul 2026 | Official leaderboard | |
| 4 | | 3.75 | Apr 2026 | Jul 2026 | Official leaderboard | |
| 5 | Claude Opus 4.8 | 5.00 | May 2026 | Jul 2026 | Official leaderboard | |
| 6 | Kimi K2.6 | 6.50 | Apr 2026 | Jul 2026 | Official leaderboard | |
| 7 | | 6.75 | May 2026 | Jul 2026 | Official leaderboard | |
| 8 | | 7.00 | Feb 2026 | Jul 2026 | Official leaderboard | |
| 9 | Grok 4.5 | 8.50 | Jul 2026 | Jul 2026 | Official leaderboard |
Rankings by part
We read 4 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Xuanhuan p303, Xuanhuan p404 and Court romance p505. DeepSeek-V4-Flash leads 2 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Xuanhuan p303
9 models we track have a score here.
- 1
DeepSeek-V4-Flash ↑3 1.00
- 2 GLM-5.2 2.00
- 3
DeepSeek-V4-Pro ↓2 3.00
- 4 GPT-5.6 Terra ↓1 4.00
- 5
Qwen3.7-Max ↑2 5.00
- 6 Claude Opus 4.8 ↓1 6.00
- 7 Kimi K2.6 ↓1 7.00
- 8 Grok 4.5 ↑1 8.00
- 9
Gemini 3.1 Pro ↓1 9.00
Xuanhuan p404
9 models we track have a score here.
- 1
DeepSeek-V4-Flash ↑3 1.00
- 2 GPT-5.6 Terra ↑1 2.00
- 3 GLM-5.2 ↓1 3.00
- 4
DeepSeek-V4-Pro ↓3 4.00
- 5 Claude Opus 4.8 5.00
- 6
Qwen3.7-Max ↑1 6.00
- 7 Kimi K2.6 ↓1 7.00
- 8 Grok 4.5 ↑1 8.00
- 9
Gemini 3.1 Pro ↓1 9.00
Court romance p505
9 models we track have a score here.
- 1
DeepSeek-V4-Pro 1.00
- 2 GLM-5.2 2.00
- 3 GPT-5.6 Terra 3.00
- 4 Claude Opus 4.8 ↑1 4.00
- 5 Kimi K2.6 ↑1 5.00
- 6
Gemini 3.1 Pro ↑2 6.00
- 7
DeepSeek-V4-Flash ↓3 7.00
- 8
Qwen3.7-Max ↓1 8.00
- 9 Grok 4.5 9.00
Court romance p606
9 models we track have a score here.
- 1 GPT-5.6 Terra ↑2 1.00
- 2
DeepSeek-V4-Pro ↓1 2.00
- 3 GLM-5.2 ↓1 3.00
- 4
Gemini 3.1 Pro ↑4 4.00
- 5 Claude Opus 4.8 5.00
- 6
DeepSeek-V4-Flash ↓2 6.00
- 7 Kimi K2.6 ↓1 7.00
- 8
Qwen3.7-Max ↓1 8.00
- 9 Grok 4.5 9.00
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official Fiction Bench (Foreverse) leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official Fiction Bench (Foreverse) 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
- 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 (9 entries).
- Pass Models scored 9 models we track.
- Pass Newest score The newest is from July 2026, for GPT-5.6 Terra.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.6 Terra (#18).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on foreverse.app: 9 rows for models we track 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.