DeepSeek previews new AI model that ‘closes the gap’ with frontier models

Chinese AI developer DeepSeek has introduced two anticipated versions of its latest large language model, DeepSeek V4, marking a significant update from last year’s V3.2 model. The new models, known as DeepSeek V4 Flash and V4 Pro, utilize a mixture-of-experts architecture with impressive context windows of 1 million tokens, enabling them to handle extensive codebases and documents effectively. The innovative mixture-of-experts approach allows only a selected number of parameters to be activated for each task, thereby reducing inference costs. Notably, the Pro version boasts an enormous total of 1.6 trillion parameters, with 49 billion active, making it the largest open-weight model currently available. This surpasses competitors such as Moonshot AI’s Kimi K 2.6, which has 1.1 trillion parameters, and MiniMax’s M1 at 456 billion. DeepSeek claims that both V4 models are more efficient and outperform the earlier V3.2 due to significant architectural enhancements. The company asserts that its V4-Pro-Max model excels against other open-source options on reasoning benchmarks and even outperforms OpenAI’s GPT-5.2 and Gemini 3.0 Pro in specific tasks. When it comes to coding competition benchmarks, the V4 models show performance levels on par with GPT-5.4. However, the models appear to lag slightly in knowledge assessments against leading competitors like OpenAI’s GPT-5.4 and Google’s Gemini 3.1 Pro. DeepSeek acknowledges this delay, estimating a developmental gap of around three to six months compared to the forefront of AI advancements. Unlike several closed-source counterparts that support multimedia tasks, both V4 Flash and V4 Pro are limited to text processing. On a positive note, DeepSeek V4 is positioned as a more cost-effective option in the market. The V4 Flash model is priced at $0.14 per million input tokens and $0.28 per million output tokens, offering a more affordable alternative to models such as GPT-5.4 Nano and Gemini 3.1 Flash. The larger V4 Pro model follows closely with a cost of $0.145 per million input tokens and $3.48 per million output tokens, also undercutting several competitors. This announcement comes shortly after accusations from the U.S. government alleging that China has engaged in widespread intellectual property theft from American AI laboratories through numerous proxy accounts. Additionally, DeepSeek has faced claims from both Anthropic and OpenAI for allegedly replicating their AI models, raising further scrutiny around its developments.