DeepSeek Unveils AI Model Rivaling Frontier Systems

Chinese AI lab DeepSeek has released two preview versions of its latest large language model, DeepSeek V4, a highly anticipated update to last year's V3.2 model and the accompanying R1 reasoning model that made a significant impact in the AI community.

The company states that both DeepSeek V4 Flash and V4 Pro are mixture-of-experts models, each featuring a context window of 1 million tokens—sufficient for processing extensive codebases or documents within prompts. This mixture-of-experts method activates only a specific subset of parameters per task to reduce inference costs.

The Pro model boasts a total of 1.6 trillion parameters (with 49 billion active), positioning it as the largest open-weight model available. It surpasses competitors like Moonshot AI's Kimi K 2.6 (1.1 trillion), MiniMax's M1 (456 billion), and more than doubles the size of DeepSeek V3.2 (671 billion). The smaller V4 Flash model contains 284 billion parameters (13 billion active).

DeepSeek claims architectural enhancements make both new models more efficient and higher-performing than DeepSeek V3.2, nearly "closing the gap" with current leading models, both open and closed-source, on reasoning benchmarks.

The lab reports that its new V4-Pro-Max model outperforms its open-source counterparts across reasoning benchmarks and exceeds the performance of OpenAI's GPT-5.2 and Gemini 3.0 Pro on certain tasks. In coding competition benchmarks, DeepSeek states the performance of both V4 models is "comparable to GPT-5.4."

However, the models appear to lag slightly behind frontier models in knowledge-based evaluations, specifically when compared to OpenAI's GPT-5.4 and the latest Google Gemini 3.1 Pro. This gap indicates a "developmental trajectory that trails state-of-the-art frontier models by approximately 3 to 6 months," according to the lab.

Unlike many closed-source peers that support audio, video, and image generation, both V4 Flash and V4 Pro are text-only models.

A key advantage is that DeepSeek V4 is significantly more cost-effective than current frontier models. The smaller V4 Flash is priced at $0.14 per million input tokens and $0.28 per million output tokens, undercutting GPT-5.4 Nano, Gemini 3.1 Flash, GPT-5.4 Mini, and Claude Haiku 4.5. The larger V4 Pro model costs $0.145 per million input tokens and $3.48 per million output tokens, also offering lower rates than Gemini 3.1 Pro, GPT-5.5, Claude Opus 4.7, and GPT-5.4.

This launch follows by one day U.S. allegations that China is conducting industrial-scale intellectual property theft from American AI labs using thousands of proxy accounts. DeepSeek has itself faced accusations from Anthropic and OpenAI of "distilling," or essentially copying, their AI models.