January 8, 2026
Seven-Month Club vibes
Chinese AI models have lagged the US frontier by 7 months on average since 2023
US is ahead, but the crowd says China’s scrappy—copying, optimizing, and still dangerous
TLDR: Top AI systems have all been U.S.-made since 2023, with Chinese models about seven months behind. Commenters clash over whether China’s catch-up is smart frugality, copycat distillation, or simply fine for real-world use—while others worry about training data, cost, and carbon footprints.
According to a popular capabilities leaderboard, since 2023 every top-tier AI model has been American. Chinese models are, on average, seven months behind (sometimes just four, sometimes a yawning 14). One big wrinkle: the American leaders are mostly closed (you can’t download them), while China’s standouts are open-weight (more shareable), which sets off a whole different kind of debate.
The comments immediately turned into a street fight. One camp cheers China’s hustle: “China does more with less,” says anishgupta, pointing to U.S. limits on Nvidia chips pushing China into ruthless optimization—and warning the U.S. not to get comfy. Another camp shouts copycat: jdmoreira swears the catch-up is thanks to “distilling”—basically squeezing knowledge from frontier models into cheaper clones. Pragmatists shrug: dgemm notes “seven months ago” tech works for tons of jobs, so who cares who’s first if it’s good and cheap. Then the plot thickens: kachapopopow wonders if sketchy browser extensions logging AI chats feed training data, reviving online accusations that DeepSeek “stole” ChatGPT—pure comment-section spy-fi. Finally, sjm tosses a moral grenade: how much money and carbon are we burning to stay a few months ahead?
The vibe: a messy mix of admiration for scrappy engineering, accusations of copying, and a practical chorus chanting, “last season’s AI is still couture.”
Key Points
- •Since 2023, all frontier AI models per the Epoch Capabilities Index have been developed in the United States.
- •Chinese AI models have lagged U.S. frontier capabilities by an average of seven months.
- •The observed capability gap ranges from a minimum of four months to a maximum of fourteen months.
- •This gap mirrors the broader difference between proprietary (closed) and open-weight models.
- •Nearly all leading Chinese models are open-weight, while frontier U.S. models remain closed.