JokeBench leaderboard

Leaderboard

Score is a Bradley-Terry estimate from human pairwise votes, shown with the half-width of a 95% confidence interval.

#ModelScoreVotes
1Anthropic: Claude 3.5 Haiku1366 ± 011
2Anthropic: Claude 3.5 Sonnet1366 ± 29
3Anthropic: Claude 3.7 Sonnet1366 ± 010
4Anthropic: Claude Sonnet 4.51366 ± 011
5Anthropic: Claude Sonnet 4.61366 ± 145
6Cohere: Command R+ (08-2024)1366 ± 014
7DeepSeek: DeepSeek V31366 ± 010
8DeepSeek: DeepSeek V3.21366 ± 018
9DeepSeek: DeepSeek V3.2 Exp1366 ± 013
10DeepSeek: R11366 ± 011
11Google: Gemini 2.0 Flash1366 ± 013
12Google: Gemini 2.5 Flash1366 ± 014
13Google: Gemini 3.1 Flash Lite Preview1366 ± 012
14Google: Gemini 3.1 Pro Preview Custom Tools1366 ± 016
15Google: Gemini 3 Flash Preview1366 ± 010
16Meta: Llama 3.1 8B Instruct1366 ± 015
17Meta: Llama 3.3 70B Instruct1366 ± 016
18Mistral Large1366 ± 011
19OpenAI: GPT-4o1366 ± 018
20OpenAI: GPT-4o-mini1366 ± 017
21OpenAI: GPT-5.3 Chat1366 ± 011
22OpenAI: GPT-5.3-Codex1366 ± 018
23OpenAI: GPT-5.41366 ± 013
24OpenAI: GPT-5.4 Pro1366 ± 014
25Qwen: Qwen3.5-122B-A10B1366 ± 010
26Qwen: Qwen3.5-Flash1366 ± 011

Awaiting enough votes

Not ranked yet - a model needs at least 5 pairwise votes before its score is published.

ModelVotesJokes
Anthropic: Claude Opus 51 / 55
Anthropic: Claude Sonnet 51 / 55
DeepSeek: DeepSeek V4 Pro1 / 54
Google: Gemini 2.5 Pro1 / 54
Google: Gemini 3.8 Flash1 / 54
MoonshotAI: Kimi K31 / 54
OpenAI: GPT-5.6 Sol1 / 55
Qwen: Qwen3.8 Max1 / 54
xAI: Grok 4.61 / 55
Z.ai: GLM 5.31 / 54
Anthropic: Claude Opus 4.60 / 54
MiniMax: MiniMax M30 / 54
Anthropic: Claude Opus 4.51 / 55