DeepSeek Unveils Enhanced AI Model V3.1, Raises Speculations on R2s Development Translation: DeepSeek Unveils Enhanced AI Model V3.1, Raises Speculations on R2s Development

The Chinese AI startup DeepSeek has updated its flagship AI model to version V3.1 and removed any reference to the reasoning neural network R1 in its chatbot. This information was reported by the SCMP.

The company announced the release of V3.1 via WeChat. This update expands the model’s contextual window to 128,000 tokens, allowing it to retain more information during user interactions, which is roughly equivalent to a 300-page book.

The model also demonstrates high efficiency in token usage.

In the Aider Polyglot benchmark, which assesses LLMs on complex programming tasks across multiple languages, DeepSeek V3.1 outperforms Claude 4 Opus.

V3.1 strikes a balance between generation speed and quality. It encompasses 685 billion parameters and is built on a hybrid architecture that ensures high performance in dialogue, reasoning, and programming tasks.

DeepSeek has eliminated references to R1 in its deep thinking feature. The SCMP speculated that this could indicate challenges in the development of the anticipated R2 version.

On August 21, the company released an official announcement on X.

Among the key features highlighted are:

The AI startup DeepSeek garnered attention in January with the release of its reasoning-focused model R1. It demonstrated significant efficiency with limited investment, leading experts to question the necessity of billion-dollar influxes into the AI sector and the inflated nature of the industry.

As a reminder, in June, the Chinese startup began hiring interns to label medical data in an effort to enhance the application of artificial intelligence in hospitals.