Recently, I tried the GLM 4.5 model on Claude Code, and the coding capabilities of this open-source model surprised me.

Using my usual test method: asking the AI to build a Next.js CRM application with a SQLite database and connecting the frontend and backend via API routes. GLM 4.5 not only set up the v15.5.2 framework using create-next-app but also automatically configured Prisma ORM and seed data—the entire process cost just $0.50, and it could be even cheaper with Zai's subscription plan.

Key observations:

1. Code generation quality is close to Sonnet 4 level

2. Excellent response speed

3. Minor issues were resolved with simple prompts, no deadlocks

4. Significant cost advantage (screenshot of expenses included)


Someone asked about comparisons with Kimi: while I haven't tested the latest version, benchmark tests show it outperforms Kimi K2. As for whether it's worth replacing Sonnet 4? The official claims don't position it as stronger, but considering the price difference, the value is truly impressive.

For now, this seems like one of the best coding options among open-source models. A heads-up: the video is sped up, so the actual coding speed isn't as dramatic.

(Not including referral links to avoid seeming too promotional—DM if needed)