Creative writing, everyday prose, brand voice, emails, long-form drafts, and tone control. This report blends public leaderboard signals into one task-specific composite score, then highlights the practical cautions that matter before you choose a model.
Claude Fable 5
97.4
2026-06-12
Top contenders
Claude Fable 5
Model 97.4 · Confidence 100%
97.4
Gemini 3 Pro
Model 97.1 · Confidence 100%
97.1
Claude Opus 4.7 Thinking
Model 99 · Confidence 79%
96.7
Claude Opus 4.6 Thinking
Model 96.2 · Confidence 100%
96.2
Claude Opus 4.7
Model 98.3 · Confidence 79%
96.1
Full-source models
5
Average coverage
58%
Top-five spread
1.3
All ranked models
Showing all 20 models with at least one confirmed source row in this category. Models with no category source coverage are excluded. Confirmed rows are ordered by Bayesian-smoothed adjusted score; missing source rows stay n/a instead of counting as zero.
High-end prose, nuanced rewriting, and difficult creative constraints.
Caution
Premium access may be the limiting factor.
100% confirmed coverage · 100% confidence
Adjusted score
97.4
Model
97.4
Confidence
100%
Research-informed writing, structured drafts, and Google ecosystem workflows.
Caution
Keep human editing for voice, claims, and citations.
100% confirmed coverage · 100% confidence
Adjusted score
97.1
Model
97.1
Confidence
100%
Long-form creative drafting where extended reasoning and voice control matter.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
55% confirmed coverage · 79% confidence
Adjusted score
96.7
Model
99
Confidence
79%
Long-form writing, editing, and careful instruction following.
Caution
May be overkill for short social or marketing copy.
100% confirmed coverage · 100% confidence
Adjusted score
96.2
Model
96.2
Confidence
100%
Polished prose, rewrites, and editorial review with strong preference ranking coverage.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
55% confirmed coverage · 79% confidence
Adjusted score
96.1
Model
98.3
Confidence
79%
Reliable daily writing, rewriting, and tone preservation.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
55% confirmed coverage · 79% confidence
Adjusted score
96.1
Model
98.3
Confidence
79%
High-end writing tasks that benefit from slower thinking-mode revisions.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
55% confirmed coverage · 79% confidence
Adjusted score
95.9
Model
98
Confidence
79%
Natural voice and human-like editing style.
Caution
Older than the newest leaderboard leaders.
100% confirmed coverage · 100% confidence
Adjusted score
95.7
Model
95.7
Confidence
100%
Fast writing iterations, content operations, and Google ecosystem workflows.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
55% confirmed coverage · 79% confidence
Adjusted score
95.3
Model
97.3
Confidence
79%
Experimental creative writing and brand-voice generation comparisons.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
55% confirmed coverage · 79% confidence
Adjusted score
95.3
Model
97.3
Confidence
79%
Open-weight oriented writing tests and lower-cost content workflows.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
55% confirmed coverage · 79% confidence
Adjusted score
95.3
Model
97.3
Confidence
79%
Alternative writing assistant testing with strong broad text Arena coverage.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
55% confirmed coverage · 79% confidence
Adjusted score
95.3
Model
97.3
Confidence
79%
Lower-latency drafts, social copy, and high-volume editing loops.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
55% confirmed coverage · 79% confidence
Adjusted score
95.3
Model
97.3
Confidence
79%
Premium writing and editing when the latest thinking variant is not needed.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
55% confirmed coverage · 79% confidence
Adjusted score
95.3
Model
97.3
Confidence
79%
Writing that needs broad context, outlines, and multimodal references.
Caution
Preview models can change; re-check before standardizing.
100% confirmed coverage · 100% confidence
Adjusted score
95.2
Model
95.2
Confidence
100%
High-effort OpenAI writing workflows with broad text preference coverage.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
15% confirmed coverage · 63% confidence
Adjusted score
94.7
Model
98
Confidence
63%
OpenAI writing and editing workflows where broad text preference is the main signal.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
15% confirmed coverage · 63% confidence
Adjusted score
94.7
Model
98
Confidence
63%
General writing drafts, outlines, and practical rewrite workflows.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
15% confirmed coverage · 63% confidence
Adjusted score
94.7
Model
98
Confidence
63%
Qwen writing tests and cost-aware multilingual content workflows.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
15% confirmed coverage · 63% confidence
Adjusted score
94.7
Model
98
Confidence
63%
Careful long drafts and editing passes when thinking-mode behavior is preferred.
Caution
Arena writing preference data is confirmed for this row; still test voice, factual claims, and brand fit before adopting it.
40% confirmed coverage · 71% confidence
Adjusted score
94.1
Model
97
Confidence
71%
How to use it
The top model is the best blended pick for this query snapshot, but model choice should still account for price, latency, privacy, context length, tool access, safety settings, and your own benchmark prompts. Use this page to reduce the search space, then run a small evaluation on your real tasks before standardizing.
