Fiction Eval leaderboard · genre
Best AI model for cultivation / xianxia writing
July 2026 — scores recomputed using only the cultivation / xianxia scenes (84 blind matchups, three tries per model). small sample — treat as rough
Claude Sonnet 4.6 leads, and comes out ahead of #2 in 93% of re-runs. This cultivation / xianxia-only sample is small, so treat the exact order as rough and check where the ranges overlap before reading much into close neighbors.
| # | Model | Cultivation / Xianxia score (likely range) |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 1785 (likely 1716–1841) |
| 2 | GLM 5.2 | 1699 (likely 1631–1762) |
| 3 | Claude Opus 4.8 | 1637 (likely 1581–1689) |
| 4 | Grok 4.3 | 1622 (likely 1552–1678) |
| 5 | Gemini 3.1 Pro | 1462 (likely 1386–1540) |
| 6 | DeepSeek V3.2 | 1355 (likely 1287–1427) |
| 7 | Mistral Large | 1288 (likely 1215–1362) |
| 8 | Llama 4 Maverick | 1150 (likely 1095–1236) |
Claude Sonnet 4.6 on the cultivation beat scene
The ranking stones never lied. That was what Master Wen had always said. Luo Fei pressed her palm to the cold obsidian pillar and watched the light crawl upward—seventh level of Qi Condensation, same as last season, same as the season before. Around her, disciples murmured behind their sleeves. She was twenty-three. At twenty-three, stagnation was not a plateau. It was a verdict. "The stone reflects what heaven has given you," Master Wen said from his dais, his voice carrying the particular gentleness he reserved for disappointing things. "Diligence cannot manufacture talent, child." She had heard those words so many times they had become a kind of scripture. She had believed them. Then Senior Brother Cao Ming placed his hand upon the pillar beside hers. The light erupted. Ninth level. The gathered elders stirred with pleasure. Cao Ming smiled with the ease of a man receiving what was already owed to him, and Master Wen descended two steps—just two, a small breach of ceremony—to clasp his favored student's shoulder. Luo …
Read every model's take side by side on the board.
More
The full 8-model board · the July 2026 report · all head-to-heads
How it works: every pair of models is judged blind on the same scene, with the passages'
order flipped so being shown first can't sway it, by GPT-5.4 — a family that isn't on the
board, so nobody scores their own side. Each score carries a likely range; overlapping ranges
are called a tie. Slop is scored separately by a fixed checklist, not an AI. Full board,
every prompt, and the FAQ: thebookfactoryai.com/board.
Get each new board by email on the model-drop list.
Writing a book of your own? Book Factory runs the same craft checks on full manuscripts.
Writing a book of your own? Book Factory runs the same craft checks on full manuscripts.