Z AI has released its GLM-5.3 artificial intelligence model via API, specializing in defensive cybersecurity, coding and complex agentic tasks. The model ties Kimi K3 for the highest intelligence score among open weights models at 60 on the Artificial Analysis Intelligence Index, a 7 point increase over GLM-5.2. In real-world agentic work, GLM-5.3’s GDPval-AA v2 Elo rose from 1524 to 1770, ranking it second overall behind Claude Opus 5 and surpassing Kimi K3's 1668.
The model maintains 753B total and 40B active parameters with a 1M token context window and will be released under an MIT license next week. While GLM-5.3 is 19% cheaper per task than Kimi K3 at $0.68, it is 20% less token efficient than GLM-5.2, requiring 18,700 output tokens per task. Knowledge accuracy on the AA-Omniscience index improved to 14 from 4, though the hallucination rate increased to 30% from 26%. API pricing remains at $1.40 per 1M input tokens and $4.40 per 1M output tokens.
Key sources
- SOURCE@zixuanli_“GLM-5.3 delivers significant improvements in reasoning, general chat, and specialized domains such as law and finance”x.com
- SUPPORT@mweinbach“Built for coding, defensive cybersecurity, and long-horizon agentic tasks”x.com
- SUPPORT@artificialanlys“the open weights frontier is closer than ever to the proprietary frontier”x.com
- SUPPORT@jun_song“It is 5x to 10x smaller than other models putting up similar numbers”x.com
- SUPPORT@cline“beating Fable & the new DeepSeek V4-Pro 0813 from just yesterday on Terminal-Bench”x.com
- SOURCE@zai_org“Built for coding, defensive cybersecurity, and long-horizon agentic tasks”x.com
- SOURCE@zixuanli_“beginning a broader review of the model weights as we work toward a responsible open-weight release”x.com
- SUPPORT@designarena“2nd-highest-ranked open-weight model on real-world design tasks”x.com