Deepseek Unveils V3.1-Terminus: A Cheaper, Improved Hybrid AI Model
Chinese AI company Deepseek has launched Gemini V3.1-Terminus, an improved version of its hybrid AI model. The new model, like its predecessors, is subject to Chinese state censorship and can be used for propaganda purposes.
Gemini V3.1-Terminus is a significant upgrade from the previous version. It has been trained with an additional 840 billion tokens, leading to enhanced performance compared to current hybrid models from OpenAI and Anthropic. The model now reliably distinguishes between Chinese and English and avoids generating incorrect special characters. It also shows better tool usage in benchmarks, with increases in BrowseComp (from 30.0 to 38.5) and Terminal-bench (from 31.3 to 36.7) scores.
The updated model addresses issues of inconsistent language and unstable outputs from the previous version. It improves performance in using agent processes and shows better tool usage. Gemini V3.1-Terminus is based on Deepseek-V3.1, which combines 'thinking' and 'non-thinking' modes for complex and simple tasks respectively. The model is accessible via app, web, and API, and open-source weights are available on Hugging Face under a MIT license.
Deepseek's Gemini V3.1-Terminus is a notable improvement over its predecessor. With better tool usage and language handling, it offers a cheaper alternative to competitors, with token output costing $1.68 per million tokens. However, like other Chinese AI models, it remains subject to state censorship and can be used for propaganda purposes.
Read also:
- EPA Administrator Zeldin travels to Iowa, reveals fresh EPA DEF guidelines, attends State Fair, commemorates One Big Beautiful Bill
- Leaders at HIT Forum 2025: Middle Powers Key to Asia's Security
- JPMorgan Chase Announces Plans for a Digital Bank Launch in Germany's Retail Sector
- Derrick Xiong, one of the pioneers behind the drone company EHang
