Z.ai Releases GLM-5.3, Challenges Open-Weight Coding Model Leaders
In brief
- Z.ai released GLM-5.3, a 743-billion-parameter coding model with lower token consumption than its predecessor.
- GLM-5.3 scores 34.5% on Z.ai's benchmark, trailing Claude Fable 5 (39.5%) and GPT-5.6 Sol on headline tests.
- Security performance reaches 84.5% on CyberGym, flagging 2,436 vulnerabilities in open-source projects.
- Zhipu's API pricing runs roughly one-tenth of U.S. frontier rates; model weights release in two weeks.
The claims and the fine print
Z.ai says GLM-5.3 achieved 34.5% on its in-house Z.ai Code Bench at Max effort while consuming roughly 75,000 output tokens per task. The lab frames this as a win on token economy—it beats Claude Opus 4.8 in that metric—but the headline numbers tell a different story. GLM-5.3 remains behind Claude Fable 5, which reaches 39.5% at Max effort.
On other benchmarks, the gap widens. On Terminal Bench 3.0, a test of autonomous shell and tool use in real Linux environments, GLM-5.3 scored 28.3, slightly behind Fable 5 (33.7) and GPT-5.6 Sol (34.6). On DeepSWE v1.1—a benchmark for fixing real GitHub issues end-to-end—open rival Kimi K3 (67.5) and Fable 5 (69.7) both beat GLM-5.3's 66.9.
The team's own framing is candid: GLM-5.3 clears its own predecessor and some open peers, but closed U.S. models still lead the headline coding boards.
Where GLM-5.3 stands out
Security work is where the model shines. GLM-5.3 leads CyberGym at 84.5% and more than doubles GLM-5.2 on exploitation benchmarks. In a real-world test, Z.ai says the model flagged 2,436 vulnerabilities across 269 open-source projects, 1,097 of them medium-to-high severity.
The efficiency angle matters too. Z.ai focused on token consumption over raw dominance—a trade-off that cuts both ways. Fewer tokens mean lower inference costs, which appeals to teams running models at scale. But it also means the model doesn't always match the highest-ceiling performance of competitors.
Pricing and availability
Zhipu's API is priced at roughly a tenth of U.S. frontier per-token rates. For context, GLM-5.2's official rate was $1.40 in / $4.40 out per million tokens, while GPT-5.3-Codex is priced at $1.75 / $14 per million tokens. GLM-5.3 weights are set for public release in about two weeks following a safety review.
One caveat: Z.ai is a Beijing lab included on the U.S. Entity List, which means American firms cannot export controlled tech to it. That context shapes the geopolitical backdrop, though it doesn't diminish the technical achievement.
