Chinese AI startup DeepSeek enhances its flagship chatbot with a new “interleaved thinking” capability in early 2026. This advanced feature allows the model to perform multi-step research while interspersing “thinking” throughout the process, marking a significant upgrade in reasoning and search functionality.
What the New ‘Interleaved Thinking’ Does
Previously, DeepSeek’s chatbot would “think” internally before delivering a final response. The new “interleaved” mode enables dynamic reasoning between actions—such as evaluating source credibility after opening a webpage or cross-verifying information. This deep research mode activates for complex queries, mimicking human-like deliberation.
Tests show it strengthens accuracy on intricate tasks, similar to features in competitors like OpenAI’s deep research or Moonshot AI’s Kimi.
Deepseek Surge in User Engagement
The update coincides with explosive growth: DeepSeek’s mobile app sees a 90% jump in monthly active users in December 2025. This consumer-side improvement comes as analysts anticipate the company’s next-generation models.
DeepSeek, founded in 2023 and backed by hedge fund High-Flyer, continues disrupting with efficient, open-source approaches despite US chip restrictions.
Industry watchers link the chatbot enhancements to preparations for advanced models, potentially building on successes like DeepSeek-R1 and V3 series. These have challenged global leaders on benchmarks while emphasizing cost-effective scaling.
Mercedes-Benz to accelerate localisation as China revenue slump hits 2025 profits 
China Bans Dual-Use Exports Including Rare Earths to Japan Over Taiwan Remarks 
North Korea Fires Ballistic Missiles Hours Before South Korean President’s China Visit 
MindRank Advances China’s First AI-Assisted Drug to Phase 3 Trials, Slashes R&D Costs by 60% 
China Enforces 50% Domestic Equipment Rule for Semiconductor Expansion 
Chinese Firms Anchor in Dubai Free Zones to Navigate Trade Barriers and Expand Reach