tech
April 24, 2026
DeepSeek previews new AI model that 'closes the gap' with frontier models
DeepSeek says both models are more efficient and performant than DeepSeek V3.2 due to architectural improvements, and have almost "closed the gap" with current leading models, both open and closed, on reasoning benchmarks.
TL;DR
- DeepSeek has launched two new large language models: DeepSeek V4 Flash and V4 Pro.
- Both models have a 1 million token context window and use a mixture-of-experts approach for efficiency.
- DeepSeek V4 Pro has 1.6 trillion parameters, making it the largest open-weight model available.
- The models reportedly show improved efficiency and performance over DeepSeek V3.2, nearly matching leading open and closed models on reasoning benchmarks.
- DeepSeek V4 Pro is claimed to outperform open-source peers and some tasks from OpenAI's GPT-5.2 and Google's Gemini 3.0 Pro, with V4 models comparable to GPT-5.4 in coding.
- The models are noted to be more affordable than current frontier models.
- Both V4 models support text only, lacking multi-modal capabilities of some closed-source peers.
- The launch follows U.S. accusations of China stealing AI IP, and DeepSeek has faced accusations of model distillation.
Continue reading the original article