DeepSeek scales web search to 92 pages

DeepSeek’s pivot to high-volume, iterative web search signals a shift toward agentic reasoning that prioritizes factual grounding over simple retrieval speed.

• –Iterative search logic generates multiple query variations and refines terms dynamically based on intermediate results.
• –New UI itemizes individual searches, providing unprecedented transparency into the model's information-gathering process.
• –Reports of increased hallucinations suggest that synthesizing 90+ pages of context remains a key technical challenge during this testing phase.
• –Move likely previews V4’s focus on logical rigor and complex reasoning rather than just parameter scaling.
• –Open-weights availability of these features would significantly lower the barrier for sophisticated local search agents.