Model Provider
The company or platform you call to run inference.
A service that hosts models and exposes them over an API — OpenAI, Anthropic, Google, and open-weights hosts like Together, Fireworks, Bedrock or your own GPUs.
Provider choice affects more than quality: data retention terms, rate limits, regional availability, price per token and how often model names change under you.
The underrated risk with providers is not price, it is lifecycle: models get retired, and a retirement is a forced migration on someone else’s schedule. Pin versions where you can, keep the provider behind your own interface, and treat the deprecation calendar as an input to your roadmap.
The migration that was scheduled for you
A provider announces the retirement of the model your product is pinned to.
The teams that survive this calmly are the ones with an eval suite that scores any model against their real tasks. Running the new model through the suite answers "is this safe to switch" in an afternoon. The teams without one discover regressions in production.
- docsDeprecations ↗
Model and endpoint retirement with notice periods and replacement mappings. The evidence that model pinning is a dependency you have to plan around.
- blogIntroducing the next generation of Claude ↗
A first-party example of a model family positioned by capability, speed and cost, with vision input and a needle-in-a-haystack long-context evaluation.
- docsPricing ↗
The current per-token and per-service prices. The right kind of source for token economics — and the reason no cost figure should be hard-coded without an access date.
Model Routing
Sending each request to the cheapest model that can handle it.
Rate Limit
The cap on how much you may send, per minute or per day.
Vendor Lock-in
How hard it would be to leave your model provider.
Inference
Using a trained model — the part you pay for on every request.