Benchmarks › EQ-Bench

EQ-Bench

Plan, revise and write a story over eight writing turns
86.3top score · Claude Opus 5
95models ranked
56open-weights models
0–100 average chapter and final-piece rubric scorehigher is better
Oct 20, 2026last updated

How the test works

1

Task

Plan, revise and write a story over eight writing turns

  • Kinds of tasksStory planning, revision, longform writing
  • In → outText → Story
2

Run

EQ-Bench longform-writing-bench; min_p=0.1

  • Temperature0.7
3

Score

0–100 average chapter and final-piece rubric score

  • Graded byClaude Sonnet 4.6 rubric judge
  • Better isHigher
Sources: eqbench.com

EQ-Bench Longform leaderboardtop 10 of 95

Full leaderboard with filters →
ModelEQ-Bench LongformPer taskPriceSpeedSupportsContext
Claude Opus 5Anthropic
86.3unstated
—$5 / $25—1M
85.3unstated
—$10 / $50—1M
84.6unstated
—$4 / $20—1M
83unstated
—$10 / $50—1M
83unstated
—$2 / $6—500K
82.8unstated
—$1.25 / $4.25—1M
82.8unstated
—$10 / $50—1.05M
GPT-6 SolOpenAI
82.8unstated
—$2 / $10—1.05M
aAion 3.5aion-labs
82.1unstated
—$3 / $6—262K
81.8unstated
—$5 / $25—1M

Versions and boards on Model Menu

BoardModels
95
91
50
1

EQ-Bench questions

What is EQ-Bench?›

Plan, revise and write a story over eight writing turns. Created by Independent project (Samuel J. Paech).

Which model is best on EQ-Bench?›

Claude Opus 5 leads EQ-Bench Longform with 86.3, ahead of Claude Fable 5.1 (85.3) and Claude Opus 5.5 (84.6).

What is the best open-weights model on EQ-Bench?›

GLM 5.3 with 81.8, then Kimi K3 (79.6).

How is EQ-Bench scored?›

0–100 average chapter and final-piece rubric score; claude sonnet 4.6 rubric judge.

Which versions of EQ-Bench are there?›

EQ-Bench Longform (95 models), EQ-Bench Creative v3 (91 models), EQ-Bench 3 (50 models), EQ-Bench 4 (1 models).