The performance gap between American and Chinese large language models has narrowed dramatically, with DeepSeek leading the charge and crypto markets already feeling the aftershocks.
Artificial Analysis, an AI benchmarking platform, reports that leading Chinese models now trail their American counterparts by just three to nine months, a gap that was well over a year not long ago.
CEO Micah Hill-Smith says his firm’s data confirms this narrowing even after the release of breakthrough models from both sides.
DeepSeek closes the gap
The headline performer on China’s side is DeepSeek. As of May 2025, the company’s R1 0528 model tied for the world’s second-ranked lab overall, making it the clear leader among Chinese AI developers.
On the American side, OpenAI’s o3 model still holds the top spot.
Chinese large language models surged from 3% to 13% of global market share within a two-month window, according to Artificial Analysis’s Q2 2025 State of AI report.
American models still dominate by raw usage, capturing roughly 93% of LLM site visits as of August 2025.
DeepSeek reportedly trained its V3 model for approximately $5.6 million, achieving this through optimized hardware and software co-design.
Why crypto cares about AI benchmarks
AI-related digital asset prices dipped following the release of DeepSeek’s competitive models.
In one trading competition, DeepSeek’s Chat V3.1 model reportedly achieved a 126% return on a $10,000 investment over nine days.
The bigger picture for investors
If Chinese LLMs continue their climb from 13% toward 20% or beyond, expect further volatility in AI-adjacent tokens that have priced in American dominance. DeepSeek’s $5.6 million training cost is a direct challenge to the narrative that frontier AI requires billions in compute spend, which matters for tokens tied to GPU rental and decentralized training networks.
