DeepSeek has released V4-Pro and V4-Flash, two massive open-weights Mixture of Experts models with 1M token context windows. The Pro model at 1.6T total parameters becomes the largest openly available model, while both offer dramatically lower pricing than frontier models from OpenAI and Anthropic. This represents a significant step in making high-capacity AI models more accessible through both open weights and affordable API pricing.
Background
DeepSeek is a Chinese AI lab known for developing large language models that compete with major Western AI companies while maintaining open weights and accessible pricing. The AI industry has been moving toward larger context windows and more efficient model architectures like Mixture of Experts.
- Source
- Simon Willison
- Published
- Apr 24, 2026 at 02:01 PM
- Score
- 8.0 / 10