Chinese artificial intelligence lab DeepSeek has introduced two preview versions of its latest large language model, DeepSeek V4, marking a significant upgrade from last year's V3.2 model. Alongside this, the new R1 reasoning model has generated considerable excitement within the AI community.
Both models, DeepSeek V4 Flash and V4 Pro, utilize a mixture-of-experts architecture featuring context windows of up to 1 million tokens. This innovative approach allows for the efficient processing of large codebases and documents by activating only a select number of parameters for specific tasks, thereby reducing inference costs.
The V4 Pro model boasts an impressive 1.6 trillion parameters, with 49 billion active parameters, making it the largest open-weight model currently available. This surpasses competitors such as Moonshot AI's Kimi K 2.6 with 1.1 trillion parameters and MiniMax's M1 with 456 billion parameters, more than doubling the size of DeepSeek V3.2 at 671 billion parameters. The smaller V4 Flash model contains 284 billion parameters, with 13 billion active.
DeepSeek asserts that both V4 models demonstrate enhanced efficiency and performance compared to their predecessor, V3.2, thanks to architectural advancements. The company claims that they have nearly "closed the gap" with leading models in reasoning benchmarks, both open and closed source.
In terms of performance, the V4-Pro-Max model reportedly outshines its open-source rivals in reasoning tasks and competes favorably against OpenAI's GPT-5.2 and Gemini 3.0 Pro in various applications. For coding competitions, both V4 models show performance levels comparable to GPT-5.4.
However, there is a slight lag in knowledge tests, particularly against OpenAI's GPT-5.4 and Google's Gemini 3.1 Pro, suggesting a developmental delay of approximately 3 to 6 months compared to the cutting-edge models.
Importantly, both V4 models are text-only, unlike many of their closed-source counterparts that support multimedia processing.
In terms of cost, DeepSeek V4 is positioned as a more affordable option compared to existing frontier models. The V4 Flash model is priced at $0.14 per million input tokens and $0.28 per million output tokens, offering a competitive edge over models like GPT-5.4 Nano and Gemini 3.1 Flash. The V4 Pro model is similarly priced at $0.145 for input tokens and $3.48 for output tokens, also undercutting several leading models.
This launch signifies a notable advancement in AI technology, suggesting that as these models evolve, they may further influence the landscape of artificial intelligence and its applications in various fields.