By task

The evaluation that matches the job, with its tested settings and source.

Top 15 · EQ-Bench Creative Writing v3

Elo rating; axis zoomed to separate the leaders. Ratings have no true zero.

ProprietaryOpen weights
180019002000210022001GPT-6 Astra21732Claude Fable 5.121623Claude Opus 521324GPT-6 Sol21255Kimi K320826GLM 5.320757Claude Opus 5.520508Grok 4.720079GPT-5.6 Sol197110ox-alpha197111Claude Fable 5194312Muse Spark 1.1192613Claude Opus 4.7191414Hemmingway-1190915Muse Spark 1.31909

1–10 of 138 · EQ-Bench Creative Writing v3

#Model / configurationElo ratingRelative percentileEvidenceDetails
1OpenAI: GPT-6 Astra

gpt-6-astra · publisher default settings

2173.1100.0llm judged
Metrics & source

rubric: 16.80

Publisher results ↗
2Anthropic: Claude Fable 5.1

claude-fable-5-1 · publisher default settings

2161.899.3llm judged
Metrics & source

rubric: 16.95

Publisher results ↗
3Anthropic: Claude Opus 5

claude-opus-5 · publisher default settings

2132.398.5llm judged
Metrics & source

rubric: 17.07

Publisher results ↗
4OpenAI: GPT-6 Sol

gpt-6-sol · publisher default settings

2124.597.8llm judged
Metrics & source

rubric: 16.51

Publisher results ↗
5MoonshotAI: Kimi K3

kimi-k3 · publisher default settings

2082.097.1llm judged
Metrics & source

rubric: 16.85

Publisher results ↗
6Z.ai: GLM 5.3

GLM-5.3 · publisher default settings

2074.896.4llm judged
Metrics & source

rubric: 17.04

Publisher results ↗
7Anthropic: Claude Opus 5.5

claude-opus-5-5 · publisher default settings

2050.095.6llm judged
Metrics & source

rubric: 16.79

Publisher results ↗
8SpaceXAI: Grok 4.7

grok-4.7 · publisher default settings

2007.494.9llm judged
Metrics & source

rubric: 17.16

Publisher results ↗
9OpenAI: GPT-5.6 Sol

gpt-5.6-sol · publisher default settings

1971.194.2llm judged
Metrics & source

rubric: 16.78

Publisher results ↗
10ox-alpha

ox-alpha · publisher default settings

1970.793.4llm judged
Metrics & source

rubric: 16.89

Publisher results ↗

About this evaluation

EQ-Bench Creative Writing v3 · Creative Writing v3

LLM-judged creative writing, not human preference or general editing accuracy. Fit uses a relative percentile within this full published pool; it is not an accuracy percentage.

Evaluated: Not published per result. Scan date records when we retrieved the source, not when these systems were tested.

138 tested configurations · 77 catalogue matches · scanned 2026-09-24. Original results ↗

Ordered by Elo rating. Fit uses the percentile within the full published pool.