Z ai released GLM-5.2, which now leads the Artificial Analysis Intelligence Index with a score of 51. While it maintains the same size as its predecessor—744B total and 40B active parameters—it outperforms rivals like Mini Max-M3 and Deep Seek V4 Pro. This update specifically targets scientific reasoning and agentic performance.

Performance gains are most evident in scientific reasoning, with a 16-point jump on Crit Pt and 12 points on HLE. The model also expands its context window to 1M tokens, a massive increase from the 200K limit in GLM-5.1. These changes allow for longer-horizon agent trajectories and better handling of complex, real-world tasks.

Efficiency remains a trade-off, as the model consumes 43k output tokens per task, making it less token-efficient than its peers. However, it sits on the Pareto frontier for cost versus intelligence, priced at $0.46 per task. It matches proprietary performance levels, specifically equaling GPT-5.5 on the GDPval-AA v2 metric.

Availability extends beyond the first-party API to providers like Deep Infra and Fireworks. The model uses an MIT license, ensuring broad accessibility for developers. GLM-5.2 achieves a lower hallucination rate of 28.1%, improving its overall reliability for technical applications.