The BYD Moment for AI
I was scrolling X before work last week, coffee in hand. z.ai dropped glm-5.1 to coding plan subscribers. No blog post, no press release. Just an X post.
The same thing happened with glm-5 turbo a few weeks earlier. I waited a week. Mid-tier plan, not the top one. Still, the wait was short.
I'm a subscriber. I switched my coding agents to glm-5.1 that morning. A few times I had to double-check which model I was using. It feels like Anthropic or OpenAI. For a fraction of the price.
Last November, z.ai ran a Black Friday deal. Mid-tier plan, about $10/month, paid yearly. Same as GitHub Copilot, but more usage. Back then, glm-4.7 was clearly worse than Claude or GPT. A gamble.
Five months later, it's the smartest $10/month I spend. And I spend some dumb $10/months.
I can't compare glm-5.1 to Claude directly. I'm not a subscriber. I can't afford it. But the benchmarks are a photo finish.
z.ai's landing page shows Claude Opus 4.6 scoring 47.9 on coding tasks. glm-5.1 scores 45.3. Same evaluation, 113 tasks, using Claude Code as the harness. 94.6% of the leader.
A few points. Not a different league.
WaveSpeedAI confirms it across more benchmarks. glm-5.1 at 77.8% SWE-bench Verified vs Claude Opus 4.6 at 80.8%. Three points, on Claude's own evaluation framework.
I don't feel much difference day to day. Maybe 5% better and 5% worse feel the same when you're in the middle of using something. Responses are faster than before. Faster is faster.
The gap used to be a canyon. Now you need benchmarks to see it.
The pace is accelerating too. Kimi K 2.5 and Minimax M2.7 dropped strong models early in 2025. z.ai released glm-5 in February, glm-5 turbo in March, glm-5.1 by end of March. Three drops from the same lab in two months. The version numbers blur.
Each one shipped within days, not teased for Q3. That pace matters more than any single benchmark score.
Years ago, a Chinese car brand called JAC landed in Mexico. Cheap, unreliable, nobody took it seriously. It was treated like a joke.
Now I look out my window in Mexico City and the streets are full of Chinese EVs. BYD, MG, Omoda. Customers seek them out. The reputation flipped so completely it's hard to remember the joke phase. Everyone assumed Chinese brands would be forever behind.
They're eating the car market.
Is it too soon to say the same will happen with AI? For cars it took years. AI moves faster. "They'll always be behind" has been wrong before. Recently. Visibly. In a market I can see from my window.
Independent developers, people paying from their own pockets, can't afford $100+/month for frontier AI. They still need to compete. Chinese labs are changing that. $10/month instead of $100+ isn't a compromise. It's a door that used to be locked.
The difference between watching from the sidelines and playing.
Solo devs can compete with bigger teams now. Not by doing the same thing faster, by getting their hands on tools they couldn't afford before. That's good for everyone, even the people paying premium prices.
More players, better game.
If you can afford premium, fine. But try the alternatives.
They're not alternatives anymore.
