Routing Across Models When a Provider Hits Capacity

Claude hitting capacity this past weekend was a useful reminder that a single-provider setup is a single point of failure. My usage limits in Antigravity drained faster than usual, fast enough that I kept switching to Gemini 3.1 Pro to keep moving. Opus 4.6 is still the most reliable model I've used for coding and it's the foundation for most of what runs on Trending Society, but "most reliable" and "always available" are not the same property, and if you ship on this stuff you have to design for the second one.

So I treat model routing as infrastructure, not a fallback you bolt on after the first outage. A second model stays wired up and ready, which turns a provider rate limit into a slowdown instead of a stop. There are leaks about a new Anthropic model, Claude Mythos, so the lineup will keep shifting, and that's exactly the point. The provider you route to will change. Whether you can route at all is the part you build once and reuse on every project.