DeepSeek launches V3.1-Terminus AI model with enhanced agent capabilities

Chinese artificial intelligence startup DeepSeek has launched DeepSeek-V3.1-Terminus, an updated version of its foundation model, just two months after releasing its previous iteration, V3.1. The new upgrade focuses on refining AI agent capabilities, addressing bugs, and improving performance based on user feedback.

Key Improvements

• Enhanced Coding & Search – The model boasts improved programming assistance and more accurate search functionalities.
• Language Consistency Fixes – Earlier versions occasionally produced mixed Chinese-English outputs or illegible symbols, which have now been corrected.
• Agent-First Approach – DeepSeek describes this update as a strategic step toward the 'Agent Era,' aiming to enable AI systems that autonomously assist users in task automation.

Benchmark Performance

According to DeepSeek’s self-reported metrics, V3.1-Terminus demonstrates slight improvements on multiple industry-standard AI benchmarks, though independent verification is pending.

DeepSeek’s Rapid Development

• The Hangzhou-based company continues to push innovation rapidly, as V3.1 was already labeled its 'most advanced model' by Artificial Analysis, a leading AI benchmarking firm.
• The push toward AI agents aligns with broader industry trends, where AI is evolving beyond chatbots into proactive problem-solving tools.

Watch related webinar: A deep dive into China’s open-source AI revolution

What’s Next?

DeepSeek’s focus on automation and AI agents positions it competitively in China’s growing AI sector. Future updates may further refine real-world task delegation, making AI copilots more seamless for users worldwide.