Chinese models are closing the gap, and I am already using them

The Washington Post ran a piece about Chinese AI models becoming a powerful source of competition for Silicon Valley. The core point: the gap is shrinking fast, and the cheapest models from Chinese labs now rival or beat frontier U.S. models on real-world tasks.

I have been using DeepSeek V4 for a lot of my headless projects — things that run in the background or as part of an agentic loop. The output is very good, especially after a bit of prompt tuning. Cheaper models are more sensitive to prompt quality, but I think that is a fair trade for running at a fraction of the cost. For a lot of work, the U.S. models are overkill and overpriced.

What this means: The benchmark gap is closing, but the bigger story is the price gap. If you can get equivalent quality for 1/10th the cost, many startups and side projects will switch without thinking. I already did.

What I am watching: The upcoming Kimi K3 model. I am interested to see if it pushes the frontier further, but also if it starts a trend of Chinese models getting more expensive. Right now, the value is in the pricing, and if that erodes, the advantage narrows. But for now, I am getting a lot of mileage out of DeepSeek and saving money.