Roleplay rankings
Which models write the best characters, and what each request costs. Scores come from EQ-Bench; prices from live routes.
| # | Model | Writing Elo | Speed | Cost / request |
|---|
Sources and methodology
Quality uses Creative Writing v3 or EQ-Bench 4. These measure story generation and social intelligence respectively; they are signals for roleplay, not complete character-chat evaluations. Scores link to their source.
Speed comes from Artificial Analysis. Providers and reasoning settings vary. Model pages retain exact settings and source dates. Missing positively weighted metrics leave a model unranked.
All-round defaults: quality 70, price 20, speed 10. Writing is scaled from Elo 200–2200 and EQ from 800–1400, clipped to 0–100. Price utility = 100 ÷ (1 + request cost ÷ $0.01). Speed utility = 100 × tok/s ÷ (tok/s + 80). The score is their weighted mean; filters never change the scale.
Snapshot: September 15, 2026. Data coverage and dates