DeepSeek V4 Preview offers a cost reduction of approximately 85% compared to GPT-5.5. The ongoing AI competition spans various domains, including those between leading AI labs and nations. In this competitive landscape, China’s DeepSeek has unveiled a glimpse of its latest model, DeepSeek V4. Unlike its American counterparts, DeepSeek is open-source, accessible under an MIT license, allowing for wider public use and modification.
The open-source model is released in two variants: DeepSeek-V4-Pro, with a 1.6-trillion parameter model, and DeepSeek-V4-Flash, with 284-billion parameters. This marks a significant step in retaining China’s prominence in open-source AI tech.
In terms of performance, DeepSeek V4 demonstrates considerable progress in task automation and coding, rivaling recent advancements from OpenAI, Google, and Anthropic. Initial benchmarks suggest that DeepSeek V4 yields comparable results to other leading AI models.
Pricing is a crucial advantage for DeepSeek V4, as it is substantially more affordable than competitors. For API usage, DeepSeek costs $1.74 per million input tokens and $3.48 per million output tokens, while OpenAI’s GPT-5.5 and Anthropic’s models have significantly higher rates. This points to DeepSeek’s strategic positioning as a cost-effective AI model, particularly beneficial for widespread adoption.
In conclusion, DeepSeek continues to challenge the status quo, presenting formidable competition with its economic and open-source models, potentially reshaping AI market dynamics.