DeepSeek has just launched the V3.1 model version with the hybrid reasoning mode 'Think & Non-Think', allowing users to choose between two ways of thinking.
V3.1 improves reasoning speed, enhances tool usability, and multi-step processing, along with a 128KB context-supporting API and standard Anthropic format, while also publicly announcing open-weight source code.
MAIN CONTENT
Version V3.1 is the first to apply the hybrid reasoning mode 'Think & Non-Think'.
Reasoning performance is faster and enhances multi-step task processing.
The expanded API supports a 128KB context, public open-weight source code, and prices are adjusted from September 5.
What highlights does the DeepSeek V3.1 model have compared to the previous version?
V3.1 introduces an integrated hybrid reasoning mode combining the 'Think' and 'Non-Think' methods, allowing users to switch flexibly via the 'DeepThink' button.
This version increases the reasoning speed compared to DeepSeek-R1-0528, while also improving the usability of support tools and executing complex multi-step tasks with higher accuracy and performance.
Notably, the version also adds an API supporting context expanded to 128KB, compatible with Anthropic's API format and allows strict function calls to enhance application capabilities in various fields.
What are the specific features and improvements of the V3.1 API?
The V3.1 API has expanded context to 128KB, a significant advancement that helps the model process longer information while maintaining a broad context for applications requiring complex analysis.
Compatibility with the standard Anthropic API format and support for strict function calls enhances system integration capabilities, expanding applications across diverse platforms, especially in big data processing and artificial intelligence.
Additionally, the open-source model weights published help the development community easily test, optimize, and apply for various needs.
"Version V3.1 marks a new advancement in the reasoning methods of artificial intelligence, helping to optimize the balance between speed and accuracy."
DeepSeek R&D Director, 2023
How are the pricing and release policy for the V3.1 model defined?
DeepSeek will adjust the usage price of the V3.1 model starting from September 5, 2023, to align with the enhanced features and upgraded performance.
The announcement of the open-weight source code provides external developers with more options to use this version, while fostering community development and increasing product transparency.
The new pricing policy aims to balance market demand and development costs, towards stability and sustainability for both providers and users.
Frequently Asked Questions
What is the most special feature of DeepSeek version V3.1?
V3.1 is the first model to integrate the hybrid reasoning mode 'Think & Non-Think', improving speed and the ability to handle complex tasks.
How many KB context does API V3.1 support?
V3.1 supports context up to 128KB, enabling much longer and deeper information processing.
Is the V3.1 model weight open-sourced?
Yes, DeepSeek has made the model weights public to help the development community easily access and customize.
When does the usage price for the V3.1 model change?
Prices will be adjusted starting September 5, 2023, to reflect the upgrades and new features.
How can the 'Think & Non-Think' mode be switched?
Users can switch between two reasoning modes using the 'DeepThink' button, suitable for different processing needs.
Source: https://tintucbitcoin.com/deepseek-v3-1-ho-tro-proxy-suy-luan/
Thank you for reading this article!
Please Like, Comment, and Follow TinTucBitcoin to stay updated with the latest news about the cryptocurrency market and not miss any important information!