DeepSeek AI Unleashes Blazing-Fast V3-0324 Model, Taking the Fight to OpenAI

Artificial intelligence (AI) has improved a lot, but challenges remain. Large language models (LLMs) need powerful computers and high costs, making them hard to use for many. These models must also have strong reasoning skills and work well on regular computers.

DeepSeek AI has tackled these problems by launching DeepSeek-V3-0324, an upgraded version of its V3 model. This model runs at 20 tokens per second on a Mac Studio, a consumer-grade device. This speed increase makes DeepSeek a stronger competitor to OpenAI.

DeepSeek-V3-0324 includes many improvements. It performs better in reasoning tests, with scores increasing:

• MMLU-Pro: 75.9 → 81.2 (+5.3)

• GPQA: 59.1 → 68.4 (+9.3)

• AIME: 39.6 → 59.4 (+19.8)

• LiveCodeBench: 39.2 → 49.2 (+10.0)

These results show that the model understands complex tasks better. It also writes cleaner code and creates better websites and game interfaces. Its Chinese writing skills have improved, and function-calling errors have been reduced.

DeepSeek AI has released DeepSeek-V3-0324 under the MIT License. This open-source approach lets developers use and modify the model without restrictions. The model’s ability to run at 20 tokens per second on a Mac Studio makes it more accessible. It reduces the need for costly hardware, helping more users benefit from AI.

With this release, DeepSeek AI takes a big step forward in AI development. By improving speed, reducing costs, and making AI more available, DeepSeek is emerging as a strong competitor to OpenAI. Its advancements will help bring AI to more people and industries.

📌Source : marktechpost