Select Language:
On August 21st, DeepSeek announced the official release of DeepSeek-V3.1, highlighting its utilization of the UE8M0 FP8 Scale parameter precision. Significant updates were made to the tokenizer and chat templates in this new version, marking clear differences from its predecessor, DeepSeek-V3. The company’s official Weibo account clarified that the UE8M0 FP8 technology is specifically designed for the upcoming generation of domestic Chinese chips, indicating a strategic alignment with future hardware developments.
In addition, to address user inquiries about the version number, the company stated that all current platforms—including their website, app, mini-program, and API open platform—have been synchronized with the latest model, which now self-identifies as DeepSeek-V3. This update demonstrates the company’s efforts to keep users informed and maintain consistency across their services.
ChatGPT
Perplexity
Gemini AI
Grok AI
