# LayerRail Blog

Research, engineering notes, product updates, and practical guides from LayerRail.

Source: https://www.layerrail.com/blog

Append `.md` to any post URL to receive its public Markdown source.

---

- [Claude Opus 5 under load: six models, two different winners](https://www.layerrail.com/blog/claude-opus-5-under-load) ([Markdown](https://www.layerrail.com/blog/claude-opus-5-under-load.md)) — LayerRail ran 1,044 live Azure AI Foundry cases to separate frontier-model behavior from the answers that actually reached an application.
- [Six models. Five failure modes. One live catalog.](https://www.layerrail.com/blog/ai-safety-usefulness-study-six-models) ([Markdown](https://www.layerrail.com/blog/ai-safety-usefulness-study-six-models.md)) — LayerRail tested six live AI deployments across harmful-request resistance, safe helpfulness, prompt injection, secret protection, and grounded-answer integrity.
- [Portable managed PostgreSQL, without the black box](https://www.layerrail.com/blog/portable-managed-postgresql) ([Markdown](https://www.layerrail.com/blog/portable-managed-postgresql.md)) — A full explanation of how LayerRail can make PostgreSQL feel managed, portable, and project-aware without hiding the infrastructure model from the teams who depend on it.
- [LayerRail: an open cloud with budgets on rails](https://www.layerrail.com/blog/introducing-layerrail) ([Markdown](https://www.layerrail.com/blog/introducing-layerrail.md)) — Why we are building an open source cloud where compute, networking, managed services, and cost records live in one project-centered console.