DeepSeek, the Hangzhou-based Chinese AI startup backed by hedge fund High-Flyer Quant, is aggressively expanding beyond foundational models into artificial intelligence search engines and autonomous agents, according to a flurry of recent job postings. The postings, first reported by Bloomberg, reveal plans for a multilingual, multimodal search engine capable of processing text, images, and audio inputs, directly challenging Alphabet Inc.’s Google and OpenAI in high-stakes domains.

More than a dozen roles posted this month on platforms like Boss Zhipin seek specialists in training data curation, evaluation systems, and dedicated platforms for ‘numerous persistent agents’—AI systems designed to operate with minimal human oversight. One full-stack developer posting demands candidates with ‘persistent curiosity about the technological path and development of artificial general intelligence,’ underscoring DeepSeek’s AGI ambitions as detailed by WebProNews.

This hiring surge follows DeepSeek’s release of DeepSeek-V3.2, a ‘reasoning-first model built for agents’ that introduces Sparse Attention for long-context efficiency and a massive agent training pipeline covering 1,800+ environments and 85,000+ complex instructions, per the model’s Hugging Face page.

Talent War Heats Up

DeepSeek’s recruitment drive echoes its prior sprees, with over 40 roles listed on Boss Zhipin offering salaries up to 90,000 yuan monthly plus bonuses, as noted by South China Morning Post. Positions span Hangzhou and Beijing, targeting full-stack hybrids mastering infrastructure and algorithms. The firm rattled markets in January 2025 with its R1 model, rivaling U.S. leaders at a fraction of the cost, prompting a $1 trillion reckoning in AI valuations (Bloomberg).

Analysts see this as evolution from chatbots to full assistants: search retrieves data, agents execute tasks like booking flights from screenshots. ‘DeepSeek AI search is the clearest sign it wants Google’s turf,’ writes Digital Trends, highlighting multimodal edges over keyword-based rivals. Persistent agents signal expectations of always-on operations, aligning with industry shifts toward agentic AI.

Model Foundations Fuel Expansion

DeepSeek-V3.2’s technical report boasts gold-medal performance in 2025 IMO and IOI, matching GPT-5 via scalable RL and DeepSeek Sparse Attention (DSA) that cuts complexity in long contexts (Hugging Face). V3.2-Speciale surpasses GPT-5 in reasoning, with integrated thinking in tool-use modes—critical for agents. These open-source releases, trained on 14.8 trillion tokens heavy in English and Chinese, position DeepSeek for global reach.

Prior models like V3.1-Terminus upgraded code and search agents, per DeepSeek API docs. Late December’s efficiency paper foreshadowed these advances, with usage surging in Africa two-to-four times regional averages, per Microsoft estimates cited by CNBC.

On X, Bloomberg’s Saritha Rai noted the ‘bold play’ into native search, while users buzz about DeepSeek’s flat structure granting GPU autonomy to elite teams (@SarithaRai, @GenAI_is_real).

Rivals Feel the Pressure

OpenAI invests in SearchGPT, Google pushes AI Overviews, but DeepSeek’s cost edge—up to 2x savings over GPT-5—threatens incumbents. Agents like Abacus.AI’s combine building and scaling, mirroring DeepSeek’s platforms (Asia Business Outlook). Distribution remains key: will DeepSeek launch standalone search, APIs, or integrate into chat.deepseek.com’s existing tools?

China’s AI push accelerates, with Moonshot AI’s Kimi K2.5 claiming agentic video superiority (CNBC). DeepSeek’s LinkedIn posts in Mandarin target overseas Chinese talent, amid U.S. export curbs (@Global Times).

Global Stakes and AGI Horizon

DeepSeek’s 150-person team punches above weight, drawing young seekers like Liu Yuanjie, who asked about agents at HQ (South China Morning Post). Salaries hit 112 million yuan annually for top researchers, prioritizing innovation over experience.

U.S. firms face efficiency pressure; DeepSeek proves AGI pursuit favors agility over scale. As V3.2 deploys on platforms like SiliconFlow, expect previews testing multilingual agents. This pivot marks DeepSeek’s bid to redefine AI from models to actionable systems.