MiniMax's Hailuo 02 Outperforms Google Veo 3 in Benchmarks
MiniMax's Hailuo 02 Surpasses Google Veo 3 in Performance Tests
MiniMax has unveiled Hailuo 02, the second-generation version of its video artificial intelligence model, showcasing significant upgrades in both performance and pricing. The new model employs an innovative architecture called Noise-Aware Computation Redistribution (NCR), which MiniMax claims improves training and inference efficiency by 2.5 times.
Technical Advancements
The NCR architecture processes long video sequences differently depending on the training stage. Early training introduces extensive artificial noise with maximum video compression, while later stages process videos at full resolution as clarity improves. Compared to its predecessor, Hailuo 02 features:
- 3x increase in parameter count
- 4x expansion of training data
- Improved data quality and diversity
Though MiniMax hasn't disclosed specific parameter counts or dataset sizes, the company reports substantial progress in handling complex prompts and simulating physical processes. Notably, Hailuo 02 is currently the only model capable of accurately generating complex scenes like gymnastics movements.
Performance and Pricing
Hailuo 02 offers three variants:
- 768p six-second video
- 768p ten-second video
- 1080p six-second video
The previous version was limited to 720p six-second videos at 25 fps. In benchmark tests by Artificial Analysis Video Arena, Hailuo 02 ranked:
- Second in image-to-video category (behind ByteDance's Seedance)
- Ahead of Google's much-anticipated Veo 3 (which currently lacks audio support)
Pricing shows significant advantages:
- API generation costs $0.28 for 768p six-second videos
- $0.49 for 1080p versions By comparison, Google's Veo 3 costs approximately $3 for eight-second 1080p videos.
Market Impact and Future Development
Since its August launch, the Hailuo platform has generated over 3.7 billion videos, attracting global creators through web interfaces, mobile apps, and APIs. MiniMax acknowledges its initial promotion was relatively random but quickly gained traction.
The company is currently working to improve:
- Generation speed
- System stability
- Additional features beyond current text-to-video and image-to-video options
The release coincides with MiniMax's launch of an open-source language model (MiniMax-M1), though technical details about Hailuo 02's training architecture remain undisclosed.
Key Points:
🌟 NCR architecture boosts efficiency by 2.5x 💰 Lower costs than competitors (Veo 3 comparison) 📊 3.7 billion+ videos generated since platform launch