744B params, tops every open benchmark, and costs just $0.80/M tokens—did Z.ai finally crack frontier performance for devs?
Imagine firing up a model that matches Gemini 3 Pro on coding and agent tasks, but it’s fully open-weights and dirt cheap. That’s GLM-5, launched February 11 by Z.ai, and it’s resetting expectations for what open models can do.
Z.ai dropped GLM-5 with a staggering 744 billion parameters (40B active), pre-trained on 28.5 trillion tokens. It hit the top spot on Artificial Analysis’ Intelligence Index at 50—the first open model to do so—while leading coding benchmarks and agentic workflows. Available now on OpenRouter at $0.80 per million input tokens, it’s not just hype; devs are already swapping it into production pipelines.[1][9]
For developers, this is huge: agentic tasks (think multi-step reasoning or tool use) now run on open models without the proprietary lock-in. No more begging for API access limits or paying premium for marginal gains—GLM-5 delivers enterprise-grade coding that beats closed rivals in speed and cost. Your next RAG app or code assistant just got 10x more viable.
Compare it to the field: DeepSeek V3.2 trails on math (93.1% vs GLM-4.7’s 95.7% AIME), and proprietary like GPT-5 still edge out on PhD science by 4-5%. But GLM-5’s MoE efficiency and agent focus make it the practical winner for most workflows, especially multilingual dev teams.[3][9]
Grab it on OpenRouter today and benchmark against your stack—what if this finally makes closed models obsolete? Watch for fine-tunes; community swarms could push it even further.
Source: AIxFunda Substack