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

Chinese AI lab DeepSeek has released two preview versions of its latest large-scale language model, DeepSeek V4. This is a long-awaited update to last year’s V3.2 model and the accompanying R1 inference model that took the AI ​​world by storm.

According to the company, both DeepSeek V4 Flash and V4 Pro are expert mixed models with a context window of 1 million tokens each, enough to use large codebases and documentation with prompts. Expert-mixed approaches activate only a certain number of parameters for each task to reduce inference cost.

With a total of 1.6 trillion parameters (49 billion active), the Pro model is the largest open-weight model available, more than twice as many as Moonshot AI’s Kimi K 2.6 (1.1 trillion), MiniMax’s M1 (456 billion), and DeepSeek V3.2 (671 billion). The smaller V4 flash has 284 billion parameters (13 billion active).

DeepSeek says both models are more efficient and performant than DeepSeek V3.2 due to architectural improvements, and have “nearly closed the gap” with current leading models in both open and closed models on inference benchmarks.

The company claims that the new V4-Pro-Max model outperforms open source equivalent models across inference benchmarks and outperforms OpenAI’s GPT-5.2 and Gemini 3.0 Pro on some tasks. In the coding competition benchmark, DeepSeek said the performance of both V4 models was “comparable to GPT-5.4.”

However, these models seem to lag slightly behind Frontier models in knowledge tests, specifically OpenAI’s GPT-5.4 and Google’s latest Gemini 3.1 Pro. The delay suggests a “development trajectory approximately three to six months behind state-of-the-art Frontier models,” the lab wrote.

Both V4 Flash and V4 Pro only support text, unlike many of their closed-source peers, which provide support for audio, video, and image understanding and generation.

Notably, the DeepSeek V4 is much more affordable than any Frontier model currently available. The smaller V4 Flash model costs $0.14 per million input tokens and $0.28 per million output tokens, lower than GPT-5.4 Nano, Gemini 3.1 Flash, GPT-5.4 Mini, and Claude Haiku 4.5. Meanwhile, the larger V4 Pro model costs $0.145 per million input tokens and $3.48 per million output tokens, which is also lower than Gemini 3.1 Pro, GPT-5.5, Claude Opus 4.7, and GPT-5.4.

The announcement came a day after the United States accused China of using thousands of proxy accounts to steal intellectual property from American AI labs on an industrial scale. DeepSeek itself has been accused by Anthropic and OpenAI of “extracting” and effectively copying their AI models.