DeepSeek is readying an artificial intelligence (AI) model with advanced agentic AI features for later this year to take on OpenAI and other rivals.
The sequel to the industry-rattling R1 model will be capable of performing more complex, multistep tasks with minimal user oversight, according to a Bloomberg report. The system is also designed to learn and improve based on prior actions.
“2026 is the year of agent monetization in China,” Sundeep Gantori, an equity strategist at UBS Global Wealth Management CIO, said in a note. As advanced models such as DeepSeek’s R2 become more sophisticated, he said, “we expect to see more monetization.”
DeepSeek is developing AI agents as part of an industrywide push toward what many consider the next evolutionary leap in AI. The upcoming release follows mounting anticipation for a follow-up to DeepSeek’s R1 model, which disrupted the global technology landscape when it debuted in January.
While the Chinese company has released only incremental updates since then, competitors across the U.S. and China have rolled out numerous new models. Reports suggest the R2’s delayed timeline reflects founder Liang Wenfeng’s perfectionist approach, though he remains focused on his profitable High-Flyer Asset Management business. Industry observers have also pointed to potential technical hurdles in the model’s development process.
The AI Agent Arms Race
As the AI agent landscape continues its rapid evolution–OpenAI, Microsoft Corp., Anthropic, and Chinese startup Manus AI each launched their own versions in recent months to streamline personal and work tasks — DeepSeek’s recent entry into the market has sparked varied reactions from industry experts.
Vernon Keenan, senior industry analyst at Keenan Vision, takes a measured approach to DeepSeek’s announcement. “Short answer, just another browser agent and not much for Anthropic or OpenAI to worry about,” he said in an email. Keenan suggests that DeepSeek’s move likely represents a browser-based agent designed to compete with OpenAI’s offerings and Claude’s Chrome plugin, rather than a revolutionary breakthrough.
His analysis highlights a critical weakness in current browser agents: Reliability issues that plague existing solutions. “Most users who have tried them once or twice have given up due to compounding errors which reduce reliability to under 90%,” Keenan said. However, he acknowledges that DeepSeek’s strong position in the Chinese consumer AI market could provide valuable training data advantages.
Keenan also points to an economic reality that benefits established players. Agents require reasoning models that consume 10 times more tokens, meaning that while per-token prices decrease, overall AI task costs may increase in the short term, potentially driving more revenue to major American model providers.
Sourcetable CEO Eoin McMillan, however, argues DeepSeek’s open-source approach could disrupt the current agent ecosystem. He suggests that while ChatGPT agents offer limited services currently, DeepSeek’s open-source nature positions it well for the multi-modal world that platforms like Hugging Face and LiteLLM are embracing.
McMillan notes the strategic timing of DeepSeek’s announcement, suggesting it “sucks attention and energy from OpenAI’s recent GPT-5 release,” highlighting the increasingly competitive nature of AI development cycles.
IBM’s Suzanne Livingston emphasizes the governance aspects of agentic AI, noting that DeepSeek’s move signals a shift in the global AI race toward systems that can “reason, decide, act, and collaborate.” But she stresses that enterprise adoption requires more than technical capability.
“When it comes to businesses, AI agents introduce great possibilities, but they also raise new questions around trust, oversight, and what’s the correct amount of autonomy,” said Livingston, highlighting IBM’s focus on building agents that are “capable, governed, observable, and aligned to company policy from day one.”
The expert consensus underscores the belief that DeepSeek’s agent initiative represents both incremental progress and strategic positioning in an increasingly crowded field.
The challenge ahead lies not just in technical capabilities, but in addressing trust, security, enterprise integration, and the economic models that will sustain this evolving ecosystem. As the AI agent arms race intensifies, success will likely depend on solving these broader systemic challenges rather than simply deploying more sophisticated models.
