GLM-5.3 leads all open models at Elo 1770 with a 19% cheaper cost per task
GLM-5.3, by China’s Z.ai, scored 60 on the top benchmark (tied with Kimi K3), achieved an Elo of 1770 in agentic tasks (second behind Claude Opus 5 at 1855), and costs $0.68/task (19% cheaper than next best).
GLM-5.3 from Z.ai China is now the #1 model globally for open weights. GLM-5.3 scores 60 on the top scoring test (tied with Kimi K3), has an Elo score of 1770 for agentic tasks (second only to Claude Opus 5 which scores 1855), and runs at a cost of $0.68/task (which is 19% cheaper than the next best model).
The open weights release was delayed 2 weeks due to GLM-5.3 being too effective at finding bugs (what I call a “cyberweapon” which cannot be released publicly). There is absolutely no doubt anymore that Chinese open source models work — they are the strongest for tasks that require thinking for yourself, rather than simply answering questions.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →▌ Related guides

Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles