DeepSeek: How Has It Disrupted Global Open Source AI?
DeepSeek AI disruption continues reshaping the global artificial intelligence landscape, as the Chinese firm and its open-source counterparts challenge long-standing US dominance by pairing near-frontier capability with dramatically lower costs.
DeepSeek recently released its latest model, DeepSeek-V4-Pro. While Chinese models are generally considered slightly behind the most advanced US systems, Chinese AI firms have been advancing rapidly. According to Artificial Analysis benchmark data, DeepSeek V4 Pro 0813 currently ranks 14th on its intelligence index, trailing models from OpenAI, Anthropic, and China’s own rising competitor Moonshot, which released Kimi K3 earlier this year.
As Kyle Chan of the Brookings Institution has noted, most Chinese models remain open-source, allowing users to customize them and access roughly 90% of a comparable US model’s capabilities at a significantly lower price. That combination has helped these models gain rapid traction among cost-conscious enterprise customers.
What Is DeepSeek?
Founded by Liang Wenfeng in 2023, DeepSeek first captured global attention in January 2025 with the release of DeepSeek R1, a model that reportedly cost far less to develop than comparable US systems. The model was widely viewed as a serious rival to OpenAI’s ChatGPT, prompting US President Donald Trump to call it a “wake up call,” urging American industries to remain laser-focused on maintaining their competitive edge.
Widely regarded as one of the most disruptive forces in AI, DeepSeek’s rise has challenged long-held assumptions about just how expensive and technically complex building state-of-the-art AI systems truly needs to be.
How DeepSeek-V4-Pro Actually Compares
According to Artificial Analysis, DeepSeek V4 Pro 0813 ranks among the leading models for raw intelligence, though it comes at a relatively higher price compared to other open-weight models of similar size, priced at $1.32 per million input tokens. By comparison, Chinese firm Z.ai prices its GLM-5.3-Flash at just $0.15 per million input tokens. This pricing strategy highlights how DeepSeek AI disruption is undercutting established market players.
Against top US models, however, DeepSeek remains considerably cheaper. Anthropic’s pricing alone varies up to tenfold across its model lineup, ranging from $0.77 per million tokens for Claude 4.5 Haiku up to $7.70 per million tokens for top-performing models like Claude Fable 5.
On the intelligence index itself, DeepSeek V4 Pro scores 57, placing it among leading models overall, though still behind the current top performer, Claude Fable 5.1, which scores 66. This composite metric tracks progress toward artificial general intelligence across mathematics, science, coding, and reasoning benchmarks.
The Broader Race Toward AGI
Like leading US firms Anthropic and OpenAI, DeepSeek has stated ambitions of eventually developing artificial general intelligence, the theoretical point at which AI systems surpass human capability across a wide range of tasks. In a 2024 interview with Chinese technology outlet 36Kr, Liang explained his approach: “Our goal is AGI, which means we need to research new model architectures to achieve stronger model capabilities with limited resources.”
Notably, OpenAI President Greg Brockman has suggested the company’s own latest model, Astra, could arguably already qualify as AGI, describing it as representing a generational leap in capability, according to reporting from the Financial Times.
The broader geopolitical competition between US and Chinese firms racing toward AGI, with relatively limited government guardrails on either side, is widely viewed as one of the more concerning dynamics in AI development today. MIT professor Max Tegmark, president of the Future of Life Institute, has been particularly blunt about the risks. “An AGI race is a suicide race,” he wrote on the institute’s website, arguing that competitive pressure leaves little room to solve unresolved technical challenges around AI control and alignment.
Open-Source AI’s Growing Global Role
The Boston Consulting Group has argued that while the US continues leading on frontier AI models, talent, and capital deployment, China is pushing hard on cost-optimized models and accelerating AI adoption across its broader economy. Chinese labs like Alibaba, Moonshot, and DeepSeek aren’t alone in this space either, with major US firms including Meta, Google, and NVIDIA all developing their own open-source models.
According to two sources familiar with internal discussions who spoke with Reuters, the Trump administration recently indicated it won’t subject open-weight AI models to voluntary safety testing, focusing that scrutiny instead on so-called “closed” models that don’t publish their underlying code. This distinction reflects broader concerns within the administration about China’s open-source momentum potentially outpacing American competitors.
Accusations Within the DeepSeek AI Disruption Story
US AI labs have repeatedly accused Chinese counterparts of a practice known as distillation, essentially using outputs from competing models to improve their own systems. In February 2026, Anthropic specifically accused Chinese AI labs DeepSeek, Moonshot, and MiniMax of running what it described as “industrial-scale campaigns” to illicitly extract Claude’s capabilities for use in their own models.
Moonshot, for its part, claims that for nine of the past twelve months, its Kimi models have set the upper bound for open-model sizes, with Kimi K3 reportedly becoming the first open model to reach 2.8 trillion parameters. Industry experts cited by the Financial Times have speculated that Anthropic’s flagship Claude Opus 4.8 likely contains somewhere between 1.5 and 2 trillion parameters by comparison.
As the broader DeepSeek AI disruption continues unfolding, the competition between US and Chinese AI labs shows little sign of slowing, with cost, openness, and raw capability all emerging as central battlegrounds in the race toward increasingly powerful AI systems.
Source: This article is based on reporting by Adam Pond for AI Magazine.