Anthropic has removed version pinning from its Claude API. Developers can no longer lock their applications to a specific model version. The change was spotted and flagged on Hacker News this week.
This is a real problem for anyone running production systems on Claude. When your app depends on stable, predictable AI outputs, you need to know exactly which model version you're calling. If Anthropic tweaks Claude's behavior and your prompts suddenly break, there's no fallback.
One HN commenter put it bluntly: "This basically makes Claude unusable for any serious production workload." Another noted they'd spent months tuning prompts for a specific version, only to find those prompts now hit a moving target.
OpenAI handles this differently. They offer dated model IDs like gpt-3.5-turbo-0301 for fixed behavior, alongside dynamic aliases that auto-update. Google's Gemini API and Azure OpenAI do the same. Developers get the choice.
Anthropic is now the lone holdout among major AI providers. No version pinning means developers absorb every change, ready or not. Anyone building systems that need reproducible results will have to look elsewhere or build fragile workarounds. Anthropic suffered a recent widespread outage, highlighting the reliability risks for developers relying on a single provider.
Anthropic hasn't publicly commented on the change.