DeepSeek previews new AI model that ‘closes the gap’ with frontier models
DeepSeek says both models are more efficient and performant than DeepSeek V3.2 due to architectural improvements, and have almost "closed the gap" with current leading models, both open and closed, on reasoning benchmarks.
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 says both DeepSeek V4 Flash and V4 Pro are mixture-of-experts models with context windows of 1 million tokens each — enough to allow large codebases or documents to be used in prompts. The mixture-of-experts approach involves activating only a certain number of parameters per task to lower inference costs.
The Pro model has a total of 1.6 trillion parameters (49 billion active), which makes it the biggest open-weight model available, outstripping Moonshot AI’s Kimi K 2.6 (1.1 trillion), MiniMax’s M1 (456 billion), and more than double 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 almost “closed the gap” with current leading models, both open and closed, on reasoning benchmarks.
The company claims its new V4-Pro-Max model outperforms its opensource peers across reasoning benchmarks, and outstrips OpenAI’s GPT-5.2 and Gemini 3.0 Pro on some tasks. In coding competition benchmarks, DeepSeek said both V4 models’ performance is “comparable to GPT-5.4.”
However, the models seem to fall slightly behind frontier models in knowledge tests, specifically OpenAI’s GPT-5.4 and Google’s latest Gemini 3.1 Pro. This lag suggests a “developmental trajectory that trails state-of-the-art frontier models by approximately 3 to 6 months,” the lab wrote.
Both V4 Flash and V4 Pro support text only, unlike many of its closed-source peers, which offer support for understanding and generating audio, video, and images.
Meet your next investor or portfolio startup at Disrupt
Meet your next investor or portfolio startup at Disrupt
Notably, DeepSeek V4 is much more affordable than any frontier model available today. The smaller V4 Flash model costs $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, meanwhile, costs $0.145 per million input tokens and $3.48 per million output tokens, also undercutting Gemini 3.1 Pro, GPT-5.5, Claude Opus 4.7, and GPT-5.4.
The launch comes a day after the U.S. accused China of stealing American AI labs’ IP on an industrial scale using thousands of proxy accounts. DeepSeek itself has been accused by Anthropic and OpenAI of “ distilling ,” essentially copying, their AI models.
When you purchase through links in our articles, we may earn a small commission . This doesn’t affect our editorial independence.
Ram is a financial and tech reporter and editor. He covered North American and European M&A, equity, regulatory news and debt markets at Reuters and Acuris Global, and has also written about travel, tourism, entertainment and books.
You can contact or verify outreach from Ram by emailing ram.iyer@techcrunch.com .
StrictlyVC kicks off the year in SF. Register now for unfiltered fireside chats and VC insights with leaders from Uber, Replit, Eclipse, and more. Plus, high-value connections that actually move the needle. Tickets are limited.
OpenAI releases GPT-5.5, bringing company one step closer to an AI ‘super app’ Lucas Ropek
OpenAI releases GPT-5.5, bringing company one step closer to an AI ‘super app’
OpenAI releases GPT-5.5, bringing company one step closer to an AI ‘super app’
Microsoft offers buyout for up to 7% of US employees Amanda Silberling
Microsoft offers buyout for up to 7% of US employees
Microsoft offers buyout for up to 7% of US employees
Duolingo is now giving users access to advanced learning content Lauren Forristal
Duolingo is now giving users access to advanced learning content
Duolingo is now giving users access to advanced learning content
Unauthorized group has gained access to Anthropic’s exclusive cyber tool Mythos, report claims Lucas Ropek
Unauthorized group has gained access to Anthropic’s exclusive cyber tool Mythos, report claims
Unauthorized group has gained access to Anthropic’s exclusive cyber tool Mythos, report claims
Tim Cook stepping down as Apple CEO, John Ternus taking over Amanda Silberling Connie Loizos
Tim Cook stepping down as Apple CEO, John Ternus taking over
Tim Cook stepping down as Apple CEO, John Ternus taking over
Blue Origin’s New Glenn put a customer satellite in the wrong orbit during its third launch Sean O'Kane
Blue Origin’s New Glenn put a customer satellite in the wrong orbit during its third launch
Blue Origin’s New Glenn put a customer satellite in the wrong orbit during its third launch
Palantir posts mini-manifesto denouncing inclusivity and ‘regressive’ cultures Anthony Ha
Palantir posts mini-manifesto denouncing inclusivity and ‘regressive’ cultures
Palantir posts mini-manifesto denouncing inclusivity and ‘regressive’ cultures
Key takeaways
- DeepSeek V4 offers a processing capacity of 1 million tokens, expanding the possibilities for AI applications in Brazil.
- The expert mixture strategy can reduce operational costs, benefiting local startups and companies.
- Despite the advancements, DeepSeek V4 still has gaps compared to top models, especially in knowledge and multimodality.
Editorial analysis
The introduction of DeepSeek V4 represents a significant advancement in the artificial intelligence landscape, especially considering the growing competition among language models. For the Brazilian tech sector, which is still maturing in relation to AI, the arrival of more accessible and efficient models can stimulate local innovations. The ability to process up to 1 million tokens at once, for instance, opens new possibilities for applications in areas such as data analysis, process automation, and software development.
Moreover, the expert mixture strategy adopted by DeepSeek could be a game changer. By activating only a subset of parameters for specific tasks, the company not only improves efficiency but also reduces operational costs. This is particularly relevant for Brazilian startups and companies looking to implement AI solutions without incurring high infrastructure costs.
However, the benchmark analysis reveals that while the new models are closing the gap with market leaders, there is still a lag in terms of knowledge and reasoning capabilities. The fact that DeepSeek V4 does not support multimodality, unlike its competitors, may limit its adoption in applications that require a richer and more integrated understanding of different types of data. This should be a point of concern for developers and companies seeking comprehensive solutions.
Finally, the accessibility of DeepSeek models, especially the V4 Flash, could democratize access to AI technology, allowing more Brazilian companies to experiment with and implement AI solutions in their operations. The reduction in costs may encourage adoption in sectors that are still hesitant to invest in AI, potentially accelerating digital transformation in Brazil.
What this coverage includes
- Clear source attribution and link to the original publication.
- Editorial framing about relevance, impact, and likely next developments.
- Review for readability, context, and duplication before publication.
Original source:
TechCrunch AIAbout this article
This article was curated and published by AIDaily as part of our editorial coverage of artificial intelligence developments. The content is based on the original source cited below, enriched with editorial context and analysis. Automated tools may assist with translation and initial structuring, but publication decisions, factual review, and contextual framing remain editorial responsibilities.
Learn more about our editorial process