AI Disruption

AI Disruption

DeepSeek Releases V3.1-Terminus Model

DeepSeek-V3.1-Terminus model released: Bug fixes, performance boosts, and enhanced Agent capabilities. Explore the latest AI update.

Meng Li's avatar
Meng Li
Sep 23, 2025
∙ Paid
4
2
Share

"AI Disruption" Publication 7800 Subscriptions 20% Discount Offer Link.


DeepSeek Releases its V3.1-Terminus Update, Featuring Improvements for  Language Consistency and Agent Upgrades - The Tech Outlook

DeepSeek Releases New Update: Is the V3 Series Coming to an End?

DeepSeek has updated again!

This evening, DeepSeek released its latest upgraded DeepSeek-V3.1-Terminus model on its official API platform (Terminus means "endpoint" or "boundary" in Latin), and announced the model's open-source release shortly after.

According to official documentation, DeepSeek-V3.1-Terminus maintains the model's original capabilities while fixing bugs that appeared after DeepSeek-V3.1's launch, including improvements to language consistency and occasional abnormal characters. It also further optimizes the performance of programming and search agents.

DeepSeek's official WeChat account published benchmark test comparisons between the old and new DeepSeek-V3.1. As shown, in non-Agent benchmarks, DeepSeek-V3.1-Terminus achieved performance improvements ranging from 0.2% to 36.5% compared to DeepSeek-V3.1, though some test scores showed minor declines.

DeepSeek-V3.1-Terminus showed the most significant performance improvement on HLE (Human Level Evaluation), which primarily tests expert-level high-difficulty knowledge and the model's multimodal and deep reasoning capabilities.

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2025 Meng Li
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture