📊 Minimax-Text-01 by Hailuo AI

Published on 1/14/2025

🚀 Minimax-Text-01 by Hailuo AI | An exciting model published on Jan 14th!

💡 Main breakthrough - Efficient super-long context handling (up to 4mil) and on-par performance with top-tier models like gpt-4o.

📊 A Quick Summary:

  • 456B model and 45.9B activated on inference
  • 100% accuracy for 4 mil token Needle-In-A-Haystack retrieval
  • Low cost low latency (api - $0.2 Mil/In, $1.1 Mil/out, 1.5s latency)
  • A cure for agentic AI memory problem

⚡ It Is:

  • Open sourced on GitHub
  • Using a new lighting attention architecture rather than traditional transformer
  • Also support multi-model (MiniMax-VL-01)

🔍 My Personal Test:

  • General: performed well, similar to claude but below gpt-o1
  • Coding: Slightly worse than Claude and Deepseek-v3 but serviceable

🤔 My thinking:

  • The most ideal model for agents that does non-coding and non-math task.

🔗 Resources: