EQ-Bench Longform Writing: leaderboard
From a short prompt the model brainstorms a plan, revises it, then writes eight chapters of roughly 1000 words each. Claude Sonnet 4.6 grades every chapter against 14 criteria, eight qualities and six flaws such as purple prose, with extra weight on forced metaphor and a penalty for overusing single-sentence paragraphs. The 0 to 100 figure averages the chapter scores and a final whole-piece score, higher is better.
Metric: Writing Score (0-100). Source: eqbench.com. Saturation forecast: Around January 2027. 140 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 86.3 | #12 |
| 2 | Claude Fable 5.1 | 85.3 | #9 |
| 3 | Claude Opus 5.5 | 84.6 | #3 |
| 4 | Claude Fable 5 | 83 | #16 |
| 5 | Grok 4.7 | 83 | #25 |
| 6 | GPT-6 Astra | 82.8 | #6 |
| 7 | GPT-6 Sol | 82.8 | #14 |
| 8 | Muse Spark 1.3 | 82.8 | #24 |
| 9 | Claude Opus 4.7 | 81.8 | #49 |
| 10 | GLM-5.3 | 81.8 | #39 |
| 11 | GPT-5.6 Sol | 81.7 | #19 |
| 12 | Muse Spark 1.2 | 81.5 | #62 |
| 13 | Claude Opus 4.8 | 80.8 | #37 |
| 14 | Claude Sonnet 4.6 | 79.9 | #90 |
| 15 | GLM-5.3 Flash | 79.9 | #65 |
No result here: #7 GPT-6.1 Sol, #8 Gemini 4 Argon, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=eq-bench-longform-writing · How It Works · Data refreshed daily, snapshot 2026-10-20.