DeepSeek Launches New AI Model Using Local Chips

Chinese artificial intelligence startup DeepSeek

BEIJING, Aug 21 (Reuters) – Chinese AI startup DeepSeek has unveiled an upgraded version of its flagship V3 model, now optimized for China-made chips and featuring faster processing speeds. The move signals a push to align with China’s growing semiconductor ecosystem amid ongoing U.S. export restrictions.

---

DeepSeek’s Rise in AI

DeepSeek made waves earlier this year by launching AI models that rival Western solutions like OpenAI’s ChatGPT but with significantly lower operational costs. The new V3.1 upgrade builds on earlier updates, including the R1 model refresh in May and the first V3 enhancement in March.

Google Launch Event 2025: Pixel 10, Pixel Watch 4, and Pixel 10 Pro Fold Announced

---

Domestic Chip Compatibility

The upgraded DeepSeek V3.1 now supports UE8M0 FP8 precision, a data format optimized for next-generation Chinese chips. While the startup hasn’t disclosed specific chip models or manufacturers, this optimization allows the AI to run faster and more efficiently while using less memory.

FP8 (8-bit floating point): Accelerates processing and reduces memory usage, giving developers and users a smoother AI experience.

---

Hybrid Inference Structure

DeepSeek’s V3.1 also introduces a hybrid inference system, letting the AI run in reasoning and non-reasoning modes.

1. Users can switch modes using the “deep thinking” button on DeepSeek’s official app and web platform.

2. The update ensures both platforms now run the latest V3.1 version, enhancing performance for all users.

---

API Access & Pricing Updates

The startup confirmed that API pricing for developers integrating DeepSeek AI into apps and web services will be updated starting September 6. This allows developers to take full advantage of the model’s new capabilities while benefiting from cost adjustments.

---

What This Means

By supporting domestic chips and improving efficiency, DeepSeek positions itself as a key player in China’s AI and semiconductor sectors. With faster processing, hybrid inference, and API improvements, the V3.1 model strengthens DeepSeek’s appeal to developers and enterprises aiming to leverage cutting-edge AI without relying on foreign technology.