DeepSeek adds interleaved thinking as users surge 90%

DeepSeek's Breakthrough Interleaved Thinking Drives User Surge to 131.5 Million

Chinese AI firm DeepSeek has introduced a significant upgrade to its chatbot, integrating an "interleaved thinking" feature that allows for more sophisticated multi-step research. This innovation marks a departure from previous models where AI would complete internal processing before generating responses.

The new capability, powered by DeepSeek V3.2 released in early December 2025, represents the company's first model to embed thinking directly into tool use. Unlike traditional approaches, the chatbot can now assess information dynamically, making crucial decisions like validating webpage credibility before seeking additional sources to confirm findings.

DeepSeek's rapid growth underscores the market impact of these advances. The platform saw monthly active users surge by 90% in December 2025, reaching nearly 131.5 million. This represents substantial growth from 33.7 million in January and 96.88 million by April of the same year, positioning DeepSeek as a major force in the global AI landscape.

How Interleaved Thinking Works

The interleaved thinking process allows DeepSeek to blend reasoning with tool execution seamlessly. When users submit complex queries, the system activates a transparent, step-by-step display showing the AI's reasoning in real-time.

This capability prevents the inefficiency of re-evaluating entire problems for each subsequent tool call. Users can observe how the AI validates sources, cross-references information, and builds comprehensive responses through iterative thinking cycles.

"This advancement places DeepSeek in direct competition with global players who are pushing the boundaries of AI reasoning capabilities," notes the company's technical documentation.

The feature particularly shines in research-intensive tasks where multiple information sources need verification. For users seeking to understand how AI reasoning models actually think, DeepSeek's transparent approach offers unprecedented visibility into the process.

Competition Heats Up in Advanced AI Reasoning

DeepSeek's rivals have developed similar functionalities. OpenAI's Deep Research, launched in February 2025 and powered by its o3 model, offers multi-step web research and comprehensive report generation. Similarly, Anthropic's Claude 4, introduced in June 2025, supports interleaved thinking capabilities.

The competitive landscape reflects broader trends in China's AI consumer war, where platforms compete fiercely for user engagement. This race for advanced reasoning capabilities highlights the ongoing evolution of AI systems beyond simple query-response interactions.

"DeepSeek V3.2 demonstrates reasoning performance on par with newer models while maintaining shorter output lengths and lower computational costs," states the company's research team.

DeepSeek's approach differs by making the thinking process visible to users. While competitors may use similar internal reasoning, DeepSeek's transparency allows users to understand how conclusions are reached, addressing concerns about AI decision-making processes explored in AI's blunders: why your brain still matters more.

Technical Innovation and Cost Efficiency

The startup has garnered attention for developing AI models that rival US counterparts at significantly reduced training costs. DeepSeek's latest technical paper, released on 31 December 2025, introduced Manifold-Constrained Hyper-Connections, signalling ongoing commitment to advancing foundational models.

This cost-effective approach resonates across Asia's AI development landscape. For organisations seeking to understand how to actually think with AI, DeepSeek's transparent reasoning process provides valuable insights into AI decision-making.

The company's innovations contribute to broader discussions about AI accessibility and development efficiency. As detailed in our coverage of DeepSeek AI: free GPT-5 rivals, the platform demonstrates how Chinese AI firms are challenging established players through innovative^✦ approaches rather than simply following existing models.

Regional Impact and Market Dynamics

DeepSeek's growth reflects broader trends in Asian AI adoption. China, India, and Indonesia account for 51.24% of DeepSeek's monthly active users, making them the platform's largest markets. This regional concentration highlights Asia's growing influence in global AI development.

The platform's success challenges assumptions about AI market dominance. While US companies have traditionally led AI development, DeepSeek demonstrates how focused innovation and cost-effective approaches can rapidly capture market share.

Key factors driving adoption include:

1. Transparent reasoning processes that build user trust
2. Cost-effective model training reducing operational expenses
3. Strong performance in mathematical and logical reasoning tasks
4. Localised optimisation for Asian markets and languages
5. Open approach to sharing technical innovations

What makes DeepSeek's interleaved thinking different from competitors?

DeepSeek provides transparent, step-by-step visibility into the AI's reasoning process, allowing users to observe how the system validates sources and builds conclusions. This transparency distinguishes it from competitors who use similar reasoning internally but don't expose the process.

How does the 90% user growth compare to other AI platforms?

DeepSeek's 90% monthly growth to 131.5 million users represents one of the fastest expansions in the AI sector. This growth rate significantly exceeds typical platform expansion rates and reflects strong user satisfaction with new features.

Can DeepSeek's interleaved thinking handle complex research tasks?

Yes, the system excels at multi-step research requiring source validation and cross-referencing. It dynamically assesses information credibility and seeks additional sources to confirm findings, making it particularly effective for comprehensive research tasks.

What are the computational costs of interleaved thinking?

Despite advanced reasoning capabilities, DeepSeek V3.2 maintains lower computational costs than comparable models while delivering shorter output lengths. This efficiency stems from the company's innovative training approaches and architectural optimisations.

Is DeepSeek's growth sustainable given increasing competition?

The platform's focus on transparency, cost-effectiveness, and continuous innovation positions it well for sustained growth. However, competition from OpenAI, Anthropic, and other major players will likely intensify as advanced reasoning becomes standard across AI platforms.