FictionBench · public tests, combined
Which AI writes the best fiction?
Paige, WriteWithPaige's own fiction model, writes your whole story and comes with every plan — every ranked model here is one you can switch to beside her. Combining 4 public fiction-writing tests, Claude Opus 5.5 ranks first (97 out of 100), ahead of GLM 5.3 and Kimi K3.1 Pro.
Updated November 18, 2026
Meet Paige
Paige
Built for fiction · included in every plan
Paige is WriteWithPaige's own fiction model — the default behind every story on the site. She reads your chapters and Story Bible, writes in your voice, keeps series details straight across a whole book, and never uses credits. Every ranked model below is one you can switch to inside the same story; Paige is the baseline you switch back to.
Knows your whole book
Chapters, characters and Story Bible are in every reply — not pasted in each time.
No credits, ever
Included in every plan with a daily allowance that refills. Ranked models use credits.
One tap to compare
Rewrite the same scene with any model below and keep the version you like.
The combined ranking
Each test ranks models its own way. We turn each rank into a position from 0 to 100 (100 means first in that test) and average them. Models need at least 2 tests to be ranked here.
| # | Model | Maker | Scoreout of 100 | Story testrank | Reader votesrank | EQ-Bench shortrank | EQ-Bench longrank |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | Anthropic | 97 | #6 of 67 | #2 of 420 | #8 of 146 | #3 of 143 |
| 2 | GLM 5.3 | Z.ai | 93 | #9 of 67 | #33 of 420 | #7 of 146 | #10 of 143 |
| 3 | Kimi K3.1 Pro | Moonshot AI | 91 | #11 of 67 | #35 of 420 | #2 of 146 | #16 of 143 |
| 4 | Gemini 4 Carbon | 87 | #19 of 67 | #3 of 420 | Not tested | Not tested | |
| 5 | Gemini 4 Argon | 83 | #27 of 67 | #1 of 420 | #32 of 146 | #14 of 143 | |
| 6 | Grok 4.7 | xAI | 81 | #32 of 67 | #85 of 420 | #9 of 146 | #4 of 143 |
| 7 | Gemini 3.8 Flash | 77 | #34 of 67 | #9 of 420 | #33 of 146 | #27 of 143 | |
| 8 | DeepSeek V4 Pro | DeepSeek | 71 | #30 of 67 | #52 of 420 | #61 of 146 | #28 of 143 |
How Paige fits in
Paige is the model you start on; every model above is one you can switch to for a second voice on a hard scene. Try the comparison yourself free — same story, same scene, both answers side by side.
Early results: the newest models
These models have only one test so far, so they don't get a combined rank yet. Their one result is shown as it stands, without a combined score.
| Model | Maker | Story testrank | Reader votesrank | EQ-Bench shortrank | EQ-Bench longrank |
|---|---|---|---|---|---|
| GPT-6.2 Astra | OpenAI | #1 of 67 | Not tested | Not tested | Not tested |
| Gemini 4 Neon | Not tested | #4 of 420 | Not tested | Not tested | |
| Claude Fable 5.2 | Anthropic | #2 of 67 | Not tested | Not tested | Not tested |
| Gemini 4 Hydrogen | Not tested | #8 of 420 | Not tested | Not tested | |
| Claude Sonnet 5.5 | Anthropic | Not tested | #32 of 420 | Not tested | Not tested |
Not in any of these tests yet: GPT-6.2 Sol and GPT-6.1 Terra.
The tests behind FictionBench
Every number below is copied from the test that measured it, with a link to the full results. Rows marked “in WriteWithPaige” are models you can pick in your own story.
Lech Mazur's story-writing test
Each model writes short stories (600 to 800 words) to the same brief, with ten story elements it must use. A panel of AI judges compares the stories head to head.
Judged by: AI judges · 67 models ranked · results from November 17, 2026 · full results
| Rank | Model, as the test names it | Score | Position0-100 |
|---|---|---|---|
| #1 | GPT-6.2 Astra (high)in WriteWithPaige | 4.302 | 100 |
| #2 | Claude Fable 5.2 (high)in WriteWithPaige | 3.564 | 98 |
| #3 | Qwen 4 Max (high) | 3.548 | 97 |
| #4 | Claude Fable 5.1 (high) | 3.51 | 95 |
| #5 | Claude Opus 5 (xhigh) | 3.443 | 94 |
| #6 | Claude Opus 5.5 (high)in WriteWithPaige | 3.355 | 92 |
| #7 | GPT-6 Astra (high) | 3.166 | 91 |
| #8 | Claude Opus 5 (high) | 3.049 | 89 |
| #9 | GLM-5.3 (max)in WriteWithPaige | 2.613 | 88 |
| #10 | Claude Fable 5 (high) | 2.585 | 86 |
| #11 | Kimi K3.1 Pro (high)in WriteWithPaige | 2.366 | 85 |
| #19 | Gemini 4 Carbon (high)in WriteWithPaige | 1.618 | 73 |
| #27 | Gemini 4 Argon (high)in WriteWithPaige | 0.454 | 61 |
| #30 | DeepSeek V4 Pro (high)in WriteWithPaige | 0.14 | 56 |
| #32 | Grok 4.7 (high)in WriteWithPaige | 0.029 | 53 |
| #34 | Gemini 3.8 Flash (high)in WriteWithPaige | -0.157 | 50 |
Text Arena creative writing
Real people type their own creative-writing prompt, read two anonymous answers and vote for the better one.
Judged by: real people, voting blind · 420 models ranked · results from November 18, 2026 · full results
| Rank | Model, as the test names it | Score | Position0-100 |
|---|---|---|---|
| #1 | Gemini 4 Argon (High)in WriteWithPaige | 1,521 | 100 |
| #2 | Claude Opus 5.5 (High)in WriteWithPaige | 1,516 | 100 |
| #3 | Gemini 4 Carbon (High)in WriteWithPaige | 1,515 | 100 |
| #4 | Gemini 4 Neon (High)in WriteWithPaige | 1,513 | 99 |
| #5 | Claude Fable 5 (High) | 1,501 | 99 |
| #6 | Claude Opus 4.6 (High) | 1,498 | 99 |
| #7 | Claude Fable 5.1 (Max) | 1,492 | 99 |
| #8 | Gemini 4 Hydrogen (High)in WriteWithPaige | 1,491 | 98 |
| #9 | Gemini 3.8 Flash (High)in WriteWithPaige | 1,489 | 98 |
| #10 | Claude Opus 4.7 (High) | 1,489 | 98 |
| #32 | Claude Sonnet 5.5 (xHigh)in WriteWithPaige | 1,460 | 93 |
| #33 | GLM-5.3 (Max)in WriteWithPaige | 1,457 | 92 |
| #35 | Kimi K3.1 Pro (Max)in WriteWithPaige | 1,456 | 92 |
| #41 | GPT-6.1 Sol (Max) | 1,450 | 90 |
| #43 | GPT-6 Astra (Max) | 1,449 | 90 |
| #52 | DeepSeek-V4-Proin WriteWithPaige | 1,446 | 88 |
| #56 | DeepSeek-V4-Pro (High, 20260813) | 1,445 | 87 |
| #85 | Grok 4.7 (xHigh)in WriteWithPaige | 1,422 | 80 |
EQ-Bench Creative Writing v3
32 writing prompts, three runs each. An AI judge grades every piece against a rubric and compares models head to head.
Judged by: AI judges · 146 models ranked · results from October 28, 2026 · full results
| Rank | Model, as the test names it | Score | Position0-100 |
|---|---|---|---|
| #1 | gpt-6-astra | 2,176.9 | 100 |
| #2 | kimi-k3.1-proin WriteWithPaige | 2,172.6 | 99 |
| #3 | claude-fable-5-1 | 2,155.3 | 99 |
| #4 | claude-opus-5 | 2,135 | 98 |
| #5 | gpt-6-sol | 2,126 | 97 |
| #6 | kimi-k3 | 2,083.3 | 97 |
| #7 | GLM-5.3in WriteWithPaige | 2,076.4 | 96 |
| #8 | claude-opus-5-5in WriteWithPaige | 2,050.9 | 95 |
| #9 | grok-4.7in WriteWithPaige | 2,012.2 | 94 |
| #10 | muse-wonder | 2,007 | 94 |
| #32 | gemini-4-argonin WriteWithPaige | 1,757.8 | 79 |
| #33 | gemini-3.8-flashin WriteWithPaige | 1,744.4 | 78 |
| #61 | deepseek-ai/DeepSeek-V4-Proin WriteWithPaige | 1,552.6 | 59 |
EQ-Bench Longform Creative Writing
Each model plans and writes an eight-chapter story. An AI judge scores every chapter and the whole story, and takes points off when quality drops over the length.
Judged by: AI judges · 143 models ranked · results from October 28, 2026 · full results
| Rank | Model, as the test names it | Score | Position0-100 |
|---|---|---|---|
| #1 | claude-opus-5 | 86.3 | 100 |
| #2 | claude-fable-5-1 | 85.3 | 99 |
| #3 | claude-opus-5-5in WriteWithPaige | 84.6 | 99 |
| #4 | claude-fable-5 | 83 | 98 |
| #4 | grok-4.7in WriteWithPaige | 83 | 98 |
| #6 | gpt-6-astra | 82.8 | 96 |
| #6 | muse-spark-1.3 | 82.8 | 96 |
| #6 | gpt-6-sol | 82.8 | 96 |
| #9 | aion-3.5 | 82.1 | 94 |
| #10 | claude-opus-4-7 | 81.8 | 94 |
| #10 | GLM-5.3in WriteWithPaige | 81.8 | 94 |
| #14 | gemini-4-argonin WriteWithPaige | 80.9 | 91 |
| #16 | kimi-k3.1-proin WriteWithPaige | 79.9 | 89 |
| #27 | gemini-3.8-flashin WriteWithPaige | 76.8 | 82 |
| #28 | deepseek-ai/DeepSeek-V4-Proin WriteWithPaige | 75.6 | 81 |
How FictionBench works
FictionBench combines public tests; it doesn't grade WriteWithPaige's own model. We pick public tests that judge whole stories, keep their results up to date, and put them side by side. People vote blind in one test; AI judges grade the stories in the other three.
Paige, our own model, is measured differently — by the writers who use her every day against every board on this page. She is included in every plan and never uses credits, so the cheapest experiment on this page is your own: start a story with Paige, rewrite a scene with any model above, keep the version you like.
A model's position in a test is where it lands among everything that test ranks: first is 100, last is 0, halfway down is 50. Its FictionBench score is the average of those positions. Big boards and small boards count the same.
We leave out tests that only count style problems such as clichés, rankings of how much a model is used, and boards with no new results since mid-2026. Tests often disagree. Gemini 4 Argon, for example, is first with readers but in the middle of the AI-judged story test. That is why one test alone can mislead.
Scores belong to the people who run each test. If you run a fiction-writing test we should include, or spot a number that has changed, email paige@writewithpaige.com.
Frequently asked questions
What is the best AI for writing fiction?
Start with Paige, WriteWithPaige's own model: she is built for fiction, knows your chapters and Story Bible, and is included in every plan. By the combined FictionBench score of public tests, Claude Opus 5.5 leads (97/100), then GLM 5.3 (93) and Kimi K3.1 Pro (91) — all available in the same story, so the best test is your own scene.
Who is Paige?
Paige is WriteWithPaige's own AI writing partner, built for fiction. She is the default model on every plan: she reads your chapters and Story Bible, writes in your voice, and never uses credits. FictionBench exists to show the outside models you can switch to beside her for a second voice.
Is Paige better than the models on this list?
Different tools, one workspace. Paige is tuned for long fiction in your own book and costs nothing extra; the ranked models are strong second opinions for a stuck scene. Every one is in WriteWithPaige, so rewrite the same scene with both and keep the version you like — that test is free.
What is FictionBench?
A leaderboard from WriteWithPaige that combines 4 public tests of fiction writing: Lech Mazur's story-writing test, Text Arena creative writing, EQ-Bench Creative Writing v3 and EQ-Bench Longform Creative Writing. We run none of these tests ourselves; we put their results side by side and link to each one. They rank the outside models you can pick beside Paige.
How is the FictionBench score worked out?
Each test ranks models its own way. We turn a model's rank in each test into a position from 0 to 100, where 100 means first and 50 means halfway down that test's list, then average the positions. A model needs at least 2 tests to get a combined rank.
Is Claude or GPT better for fiction?
On the combined score, the top Claude model is Claude Opus 5.5 (97/100). The top GPT model, GPT-6.2 Astra, is #1 in Lech Mazur's story-writing test, but only one test covers it so far. Both are in WriteWithPaige beside Paige, so you can try each on the same scene.
How much does it cost to write with Paige?
Paige is included in every WriteWithPaige plan, free ones included, with a daily allowance that refills. She never uses credits; the outside models on this board use credits, sold from US$10. You can hold the same scene against both on day one.
How often is FictionBench updated?
When a source publishes new results or a new model comes out. Each test shows the date of the results we used.
Try the models on your own story
Start with Paige. Compare with the board.
Write a chapter with Paige for free, then rewrite it with any model on this list and keep the version you like.
Start free