DeepSeek has sent shockwaves through the AI industry by announcing a permanent 75% price cut for its V4 Pro flagship model. This move is seen as a direct assault on the expensive token-based business models of Western AI giants.
Background
DeepSeek V4 Pro's input pricing is currently 7 times cheaper, and its output is 17 times cheaper, than Claude Sonnet or GPT-5.5-Med. Thanks to its Multi-head Latent Attention (MLA) architecture and hardware optimizations, DeepSeek can maintain near-top-tier performance at extremely low operating costs. The model is also available as open weights, allowing enterprises to self-host.
Why it matters
For Vietnamese enterprises building multi-agent systems, token costs have always been the biggest hurdle. The arrival of the ultra-affordable DeepSeek V4 Pro enables large-scale automated workflows (swarms) without budget concerns. It also pressures OpenAI and Anthropic to lower prices or offer more value to retain customers.