Released: 2026-07-25
Key:
Score Threshold:
openai/gpt-4o
openai
(4 * 50) + 75 - 100 + 25
= 200
google/gemini-3-flash-preview
100 + 4 * (50 - 10 - 25/75)
= 258
deepseek/deepseek-chat-v3-0324
deepseek
(75 - 25) * (10 - 4) + 50 + 100
= 450
mistralai/mistral-medium-3
mistral
(100 + 50 + 75 + 25) * (10 - 4)
= 1500
moonshotai/kimi-k2.5
moonshot
(25 * 10) - (100 / 50)
= 248
z-ai/glm-5
z-ai
(25 * 10) - (4 / (100 / 50))
= 248
minimax/minimax-m2.5
minimax
(100 * 10) / 4 - 50 / 25
= 248
anthropic/claude-sonnet-4.6
anthropic
100 * 4 - 75 + 25 - 50 / 10
= 345
Each model receives the same prompt with the numbers to use. Models are tasked with creating an expression using only arithmetic operations to reach the target number. Each number can only be used once and you do not have to use all the numbers. Their answers are evaluated without feedback or retries.