Select Language:
Chinese AI startup DeepSeek announced an upgrade to its flagship V3 model on Thursday, featuring enhanced capabilities to optimize performance with domestic Chinese-made chips and increased processing speed. This focus on support for local hardware suggests DeepSeek is aiming to align its AI models with China’s growing semiconductor industry as Beijing accelerates efforts to develop homegrown technology amidst U.S. export restrictions.
Earlier this year, DeepSeek made waves in the tech sector by launching AI models that rival Western counterparts like OpenAI’s ChatGPT, all while boasting lower operational costs. These recent updates to its core models, including the V3 enhancement in March and an R1 version in May, demonstrate ongoing evolution.
For compatibility with China’s next-generation chips, DeepSeek introduced its DeepSeek-V3.1 model, which utilizes the UE8M0 FP8 precision format—an 8-bit floating point system designed to boost efficiency by reducing memory requirements and increasing speed. DeepSeek indicated in a WeChat post that this optimization is tailored for “next-generation domestic chips” expected to debut soon, though specific models or manufacturers weren’t named.
The FP8 data format enables AI models to run more efficiently, consuming less memory and delivering faster processing—benefits that are crucial for deploying AI at scale within China’s local hardware landscape.
DeepSeek’s V3.1 also features a hybrid inference architecture allowing the model to switch between reasoning and non-reasoning modes. Users can toggle these modes via a “deep thinking” button available on the company’s official app and web platform, both now running the latest version.
Starting September 6, the company plans to modify API pricing for accessing its models—a platform that enables developers to embed DeepSeek’s AI capabilities into their own applications and websites.
ChatGPT
Perplexity
Gemini AI
Grok AI
