Chinese AI lab DeepSeek has launched two preview versions of its newest large language model, DeepSeek V4, a much-awaited update to last year’s V3.2 model and the accompanying R1 reasoning model that took the AI world by storm.
The company announced that both DeepSeek V4 Flash and V4 Pro are mixture-of-experts models featuring context windows of 1 million tokens each, allowing for the integration of extensive codebases or documents into prompts. This mixture-of-experts approach activates only a selected number of parameters per task, which helps to reduce inference costs.
The Pro model boasts a total of 1.6 trillion parameters, with 49 billion active, making it the largest open-weight model currently available. This surpasses rivals such as Moonshot AI’s Kimi K 2.6 with 1.1 trillion, MiniMax’s M1 at 456 billion, and more than doubling the 671 billion parameters of DeepSeek V3.2. The smaller, V4 Flash model contains 284 billion parameters, with 13 billion active.
DeepSeek asserts that both V4 models are more efficient and high-performing than V3.2 due to architectural enhancements, claiming they have nearly “closed the gap” with the leading models, whether open or closed, on reasoning benchmarks. The company maintains that its new V4-Pro-Max model surpasses its open-source competitors across reasoning benchmarks, even outpacing OpenAI’s GPT-5.2 and Gemini 3.0 Pro in certain tasks. In coding competition benchmarks, DeepSeek claims the performance of both V4 models is “comparable to GPT-5.4.”
Despite these advancements, the models appear to lag behind frontier models in knowledge tests, particularly OpenAI’s GPT-5.4 and Google’s Gemini 3.1 Pro. This discrepancy suggests a “developmental trajectory that trails state-of-the-art frontier models by approximately 3 to 6 months,” according to the lab.
Both V4 Flash and V4 Pro are limited to text-only capabilities, contrasting with many closed-source competitors that offer multi-modal functionalities, including audio, video, and image processing.
In terms of pricing, DeepSeek V4 is significantly more affordable than existing 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 has a cost of $0.145 per million input tokens and $3.48 per million output tokens, also underpricing competitors like Gemini 3.1 Pro, GPT-5.5, Claude Opus 4.7, and GPT-5.4.
The announcement follows a day after the U.S. government accused China of large-scale theft of American AI intellectual property through numerous proxy accounts. DeepSeek itself faces allegations from Anthropic and OpenAI regarding the “distillation,” or copying, of their AI models.
As DeepSeek enters the competitive landscape with its V4 models, the implications for its positioning in the rapidly evolving AI sector remain significant, particularly as the race for dominance intensifies amid allegations of intellectual property disputes and advancing technological capabilities.
See also
Germany”s National Team Prepares for World Cup Qualifiers with Disco Atmosphere
95% of AI Projects Fail in Companies According to MIT
AI in Food & Beverages Market to Surge from $11.08B to $263.80B by 2032
Satya Nadella Supports OpenAI’s $100B Revenue Goal, Highlights AI Funding Needs
Wall Street Recovers from Early Loss as Nvidia Surges 1.8% Amid Market Volatility
