🤖 Large Language Models

DeepSeek V4: Why the $0.04 Model Crushed Pro-Max

Did the $0.04 DeepSeek V4 model just outgun its pricier sibling? We tested 4 modes on 20 real-world tasks, and the results might shock you.

A graphic showing a comparison of AI model performance and cost, with a smaller, cheaper model leading.

⚡ Key Takeaways

  • DeepSeek V4's $0.04/million token 'Flash' mode surprisingly outperformed pricier versions on 7 of 20 real-world tasks. 𝕏
  • A "10% KV cache trick" is cited as a key architectural innovation enabling the Flash model's efficiency. 𝕏
  • The Pro-Max mode used 4.3x more tokens for a marginal 2-point performance gain, highlighting potential over-engineering and cost inefficiency. 𝕏
Marcus Rivera
Written by

Marcus Rivera

Enterprise AI correspondent. Covers how businesses adopt, fund, and operationalize AI.

Worth sharing?

Get the best AI stories of the week in your inbox — no noise, no spam.

Originally reported by Towards AI

Stay in the loop

The week's most important stories from The AI Catchup, delivered once a week.