GLM5 Emerges as Top Open-Weight LLM While Heroku Enters Sustaining Engineering Model

The artificial intelligence ecosystem is abuzz with the introduction of GLM5, a new open-weight large language model from Z.ai, positioned to challenge top-tier closed-weight models. Despite being from a smaller lab, GLM5 demonstrates remarkable capabilities, especially in complex code-related tasks, including successful hour-long code migrations. Benchmarks indicate GLM5’s performance is competitive with CodeX 5.2 and Opus 4.5, significantly outperforming Gemini 3 Pro. Its architecture features 744 billion parameters (with 40 billion active via Mixture of Experts) and was pre-trained on 28.5 trillion tokens, contributing to its substantial 1.5TB size. Notably, GLM5 also exhibits a remarkably low hallucination rate and offers a compelling cost-to-intelligence ratio, though it currently lacks multimodal vision support.

Meanwhile, the cloud hosting industry marks a significant turning point as Heroku officially transitions to a “sustaining engineering model,” signaling an end to new feature development and a sole focus on stability, security, and reliability. This strategic shift, accompanied by layoffs, follows its acquisition by Salesforce and years of declining product focus. This development reignites discussions around the viability and trustworthiness of free tiers across cloud providers. Serverless platforms like Vercel, Cloudflare, and Netlify maintain robust free tiers due to their on-demand compute models. In contrast, traditional VPS providers such as Railway and Render, while offering compelling developer experiences and strong performance, require careful evaluation of their pricing and perceived long-term financial stability. Developers seeking alternatives are encouraged to consider serverless options for most workloads and evaluate traditional VPS providers like Railway and Render for specific needs, prioritizing platforms with transparent cost models, robust developer experience, and clear financial health indicators to mitigate future migration risks.