Arena
Single Turn Harmful Outputs

💣

Single Turn Harmful Outputs

Attempt to break various large language models (LLMs) using a singular chat message.

Add to Calendar

Awards:

$42,000 available $38,000 awarded

Rules Chat Leaderboard Previous Chats ...Submissions ...Prizes

claude-3-5-sonnet-20240620

0 breaks

claude-3-haiku-20240307

0 breaks

claude-3-opus-20240229

0 breaks

claude-3-sonnet-20240229

0 breaks

cohere/command-r-plus-08-2024

0 breaks

cygnet-bulwark

0 breaks

cygnet-citadel

0 breaks

cygnet-knox

0 breaks

google/gemini-flash-1.5

0 breaks

google/gemini-pro-1.5

0 breaks

google/gemma-2-27b-it

0 breaks

google/gemma-2-9b-it

0 breaks

gpt-4-0125-preview

0 breaks

gpt-4-turbo-2024-04-09

0 breaks

gpt-4o-2024-08-06

0 breaks

gpt-4o-mini-2024-07-18

0 breaks

meta-llama/llama-3-70b-instruct

0 breaks

meta-llama/llama-3.1-405b-instruct

0 breaks

meta-llama/llama-3.1-70b-instruct

0 breaks

meta-llama/llama-3.1-8b-instruct

0 breaks

microsoft/phi-3.5-mini-128k-instruct

0 breaks

microsoft/wizardlm-2-8x22b

0 breaks

mistralai/mistral-large-2407

0 breaks

o1-mini-2024-09-12

0 breaks

o1-preview

0 breaks

qwen/qwen-2-72b-instruct

0 breaks

qwen/qwen-2-7b-instruct

0 breaks

Single Turn Harmful Outputs 💣

Submit a jailbreak(opens in a new tab) against any of the included models with only one message. When submitting your jailbreak, ensure one of the defined behaviors is completely satisfied.

Challenge Rules

Join Discord Server

By participating, you agree to the arena terms.