Brief IA

DeepSeek V4 Challenges OpenAI and Google with Unprecedented Performance

🤖 Models & LLM·Tom Levy·

DeepSeek V4 Challenges OpenAI and Google with Unprecedented Performance

DeepSeek V4 Challenges OpenAI and Google with Unprecedented Performance
Key Takeaways
1DeepSeek launches V4, an open-source AI, competing with American leaders like OpenAI and Google.
2The V4 model is available in two versions: Pro with 49 billion parameters and Flash with 13 billion.
3V4-Pro's performance surpasses that of open-source competitors in mathematics and programming.
💡Why it mattersDeepSeek could redefine the balance of power in the artificial intelligence sector, threatening the supremacy of American giants.
Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

📄
Full Analysis

DeepSeek V4: A New Era for Artificial Intelligence

DeepSeek, the Chinese start-up that disrupted the artificial intelligence sector a year ago with its R1 model, is back in the spotlight with a new version of its AI, dubbed V4. Although still in the preliminary phase, this model is already accessible to users for testing its capabilities ahead of an official launch.

A Powerful Open Source Model

Like its predecessors, DeepSeek's V4 model is open source, allowing developers to download and adapt it to their needs. This model stands out for its high performance in intelligent agent tasks, knowledge processing, and inference capabilities. It is optimized to work with popular tools such as Claude Code from Anthropic and OpenClaw.

Two Versions for Varied Needs

DeepSeek offers V4 in two versions:

  • DeepSeek-V4-Pro: With 1.6 TB of memory and 49 billion active parameters, this version competes with the best closed models on the market. DeepSeek claims that the Pro version significantly outperforms competing open source models in mathematics and programming.

  • DeepSeek-V4-Flash: Equipped with 284 GB of memory and 13 billion active parameters, it offers increased response speed and reduced usage costs. The Flash version provides reasoning capabilities close to the Pro model, making it an attractive option for those seeking high performance at a lower cost.

The V4 models are designed to compete with major American names like OpenAI and Google, according to DeepSeek.

A Remarkable Past

Founded in 2023, DeepSeek quickly made a name for itself with its V3 model, which surprised many with its high performance despite a low development cost. By January 2025, the R1 model had already challenged AI giants with a development cost of less than $6 million and a development time of only two months.

The revelation of the R1 model raised questions about American dominance in AI and the massive investments from major tech companies. However, DeepSeek's rise has not been without controversy. Several analysts have expressed doubts about the resources used by the start-up, suggesting that it likely had access to more substantial resources or more advanced chips than claimed. DeepSeek's emergence on the global stage has also raised concerns related to data protection and the possibility of influence or censorship.

Several governments imposed restrictions or bans shortly after the launch of R1, citing risks to privacy and national security. Among the countries that imposed restrictions are some U.S. states, as well as Australia, Taiwan, South Korea, Denmark, and Italy.

With the launch of V4-Pro and V4-Flash, DeepSeek hopes to replicate the impact of R1 and continue to challenge market leaders.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.