DeepSeek Model Emerges as Cheapest AI to Run, Study Finds

DeepSeek's Cost Advantage Stuns Benchmarks

A new benchmark analysis has found that a version of Chinese startup DeepSeek's flagship AI model is dramatically cheaper to run than any other well-known AI system in the world, undercutting rivals by more than 100 times in some comparisons, including Anthropic's Claude models 1. The finding, reported by a research firm tracking model performance and cost, underscores how quickly Chinese AI labs have shifted the economics of large-scale AI deployment, turning price into as much of a competitive battleground as raw capability 1.

A Broader Chinese Push on Price and Openness

DeepSeek's pricing edge is not an isolated event but part of a wider pattern of Chinese firms undercutting Western AI providers on cost while embracing open-weight distribution. Reporting on the Chinese AI sector notes that models from companies like DeepSeek and Alibaba are gaining ground globally, including inroads into the U.S. market, largely because they combine competitive performance with far lower price tags and more permissive licensing 4. Alibaba's Qwen family has become a central example of this strategy: the company is reportedly preparing to ask major commercial users of its next-generation open-source Qwen model to share a portion of the revenue they generate from it, a notable shift from the purely free-to-use image many open-source models have cultivated 3. That move suggests Chinese firms are trying to monetize scale and enterprise adoption even while keeping listed prices for compute far below their American counterparts 34.

Industry Leaders React to the Shifting Economics

The ripple effects are already shaping strategic thinking in the software industry. Box CEO Aaron Levie has warned that the rapid ascent of open-weight models like Qwen could trigger a broader collapse in AI pricing, arguing that closed, proprietary AI providers cannot indefinitely justify premium costs as open alternatives close the performance gap 6. Levie's comments reflect growing anxiety among enterprise software leaders that competitive advantage will increasingly migrate from the underlying models themselves toward the applications built on top of them, as commoditization pushes margins downward across the model layer 6.

Do Consumers Even Need the Priciest Models?

Parallel commentary aimed at everyday users argues that most people do not need to pay for the newest, most expensive AI models at all. Unless someone is doing intensive tasks like AI-assisted coding or generating high-end media, older or cheaper models paired with better-crafted prompts can deliver comparable practical results, reducing the incentive to chase the latest benchmark leader 7. Taken together with the DeepSeek and Qwen developments, this suggests a market bifurcating: a small segment of users genuinely need frontier-level capability, while a much larger group can rely on inexpensive or open models without meaningfully sacrificing utility 71.

Why It Matters

The convergence of ultra-low-cost inference, aggressive open-weight releases, and monetization experiments signals a maturing AI market where cost efficiency, not just raw intelligence, is becoming the key differentiator. As Chinese models undercut incumbents on price and open licensing spreads, established AI providers face mounting pressure to justify premium pricing, potentially reshaping competitive dynamics across the global AI industry 1346.