Stay Ahead of the Curve
Latest AI news, expert analysis, bold opinions, and key trends — delivered to your inbox.
Z.ai Drops GLM-4.5: Open-Source Agentic AI That’s Cheaper Than DeepSeek—and Nearly as Good as o3
3 min read Z.ai just launched GLM-4.5—an open-source AI model that rivals o3 and Grok 4 in performance, outperforms them in agentic tasks, and costs less than DeepSeek. China’s open frontier push is getting serious. July 29, 2025 14:28
Chinese AI startup Z.ai (formerly Zhipu) just launched GLM-4.5, an open-source family of agentic AI models designed to handle reasoning, coding, and autonomous tasks—and it's gunning for the top. With performance approaching models like o3 and Grok 4, and a price point that undercuts DeepSeek, Z.ai is making bold moves in the open frontier.
Key Highlights:
-
One model, many talents: GLM-4.5 packs 355B parameters and combines reasoning, coding, and agentic capabilities with “hybrid thinking” to balance speed and task complexity.
-
Top-tier performance: Z.ai claims 4.5 is now the best open-source model globally, placing just behind o3 and Grok 4 overall—while outperforming them in agentic benchmarks.
-
Agentic dominance: GLM-4.5 beats Gemini 2.5 Pro, o3, and Grok 4 in tool-use tasks, scoring a 90% success rate—a major leap for autonomous AI agents.
-
Fully open-source: Both GLM-4.5 and the lighter 4.5-Air launched with open weights. Z.ai also open-sourced its custom ‘slime’ training framework, inviting the community to build on it.
Why it matters
From Qwen to Kimi, MiniMax to Z.ai, Chinese labs are dropping high-quality open models at a breakneck pace. The frontier gap is shrinking fast—and companies like OpenAI are now under real pressure to keep their lead. GLM-4.5 isn’t just another model; it’s a sign of the global open-source arms race heating up.
User Comments (0)
Editor's Choice
AI Agents