Chinese AI firm DeepSeek has launched a preview of its highly anticipated V4 model, which significantly improves upon its predecessors. The new model can process longer prompts more efficiently, thanks to its innovative design. As with DeepSeek’s previous models, V4 is open source, allowing anyone to download, use, and modify it.
V4 marks a major milestone for DeepSeek, which rose to prominence with the release of its R1 model in January 2025. The R1 model stunned the global AI industry with its impressive performance and efficiency, despite being trained on limited computing resources. The release of V4 comes after months of scrutiny, including personnel departures and delays to previous model launches.
So, why does V4 matter? Here are three key reasons:
- V4 breaks new ground for open-source models, offering performance rivaling the best models available at a fraction of the cost.
- The model comes in two versions: V4-Pro, designed for complex tasks, and V4-Flash, a smaller, faster, and more affordable version.
- V4 offers reasoning modes, allowing the model to carefully parse user prompts and display each step as it works through the problem.
In terms of pricing, V4-Pro is available for $1.74 per million input tokens and $3.48 per million output tokens, while V4-Flash is priced at $0.14 per million input tokens and $0.28 per million output tokens, making it one of the most affordable top-tier models available.
Photo by Okiki Onipede on Pexels
Photos provided by Pexels