Kaggle Joke Arena: leaderboard

Kaggle's human-vote humour arena: each model answers 100 'tell me an original joke about X' prompts in 10 categories (a valid answer is a PG joke of 3-150 words), visitors vote between two models' jokes, and Kaggle fits an arena rating from the votes. Rows are model and reasoning-effort pairs; with 15 to 84 votes per model the 95% intervals span about 100 to 170 points either way, so most neighbouring ranks are not separable.

Metric: Arena rating (human votes). Source: www.kaggle.com. Saturation forecast: Not forecast. 25 models tracked. Updated 20 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Kimi K3Max1584.52#61
2Gemini 3.1 Pro (Preview)High1567.96#101
3GPT-6 AstraMax1559.57#10
4Claude Fable 5.1Max1545.29#20
5GLM-5.3Max1544.14#73
6Gemini 3.8 FlashHigh1539.06#43
7Claude Sonnet 5.5Max1533.47#24
8GPT-6.1 SolMax1521.81#15
9Claude Opus 5Max1512.68#26
10GPT-5High1511.28#180
11Claude Opus 5.5Max1509.93#9
12Claude Opus 4.7High1502.78#85
13GPT-5.6 SolMax1501.8#36
14GPT-5.5High1500.47#54
15GPT-5.6 TerraHigh1492.11#72

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=kaggle-joke-arena · How It Works · Data refreshed daily, snapshot 2026-11-20.