Chinese AI startup DeepSeek unveils v3.1 model for testing

DeepSeek has announced what appeared to be an updated version of its V3 AI model, called V3.1, and says it is ready for testing.

The Hangzhou-based AI startup said the new model has a longer context window, which enables it to process more information in each query.

DeepSeek did not share further technical details and has yet to publish documentation on major platforms such as Hugging Face.

The company’s earlier models have drawn attention for their speed and performance, competing with US firms like OpenAI.

DeepSeek’s R1 model, released earlier this year, outperformed several Western AI models in standardized tests.

The company has not yet launched its awaited R2 model, with local reports citing delays due to technical issues and CEO Liang Wenfeng’s high standards.

🔗 Source: Bloomberg

🧠 Food for thought

1️⃣ Extreme cost efficiency creates new competitive dynamics in AI

DeepSeek’s pricing model represents a dramatic shift in AI economics, with R1 operating at approximately 5% of the cost of OpenAI’s o1 series 1.

This cost advantage extends beyond just R1, with reports indicating DeepSeek’s models are 20 to 50 times cheaper than OpenAI’s offerings across their product line 2.

The cost efficiency stems from their innovative R1-Zero training method, which relies solely on reinforcement learning and chain-of-thought reasoning, potentially requiring fewer computational resources than traditional approaches 1.

This pricing disruption forces established players to reconsider their cost structures while making advanced AI capabilities accessible to budget-sensitive startups and research initiatives that previously couldn’t afford premium models.

The open-source nature of DeepSeek’s models allows businesses to deploy and customize AI solutions without incurring the high recurring costs associated with proprietary platforms 2.

2️⃣ Hardware dependency exposes vulnerabilities in China’s AI self-sufficiency push

DeepSeek’s struggles with its R2 model reveal the practical challenges of domestic technology adoption despite government pressure.