Scrapegraph-ai carries its own table of models and their context windows, and the open issues show how hard that is to keep honest: a model name that is not in the table falls back to a small window and truncates the page without saying so, the effective window is never reported back to the caller, and the table keeps needing corrections as vendors move. A scrape also burns a lot of tokens in one run, so which model sits behind it becomes a cost question before it is a quality one.
BlockRun serves chat, image, video, music, speech and sound effects behind one OpenAI-compatible base URL, and its model list can be read at runtime rather than assumed, with each call settling in USDC over x402 instead of against an API key. https://blockrun.ai
Would a provider like that be worth carrying, or does the model table already cope with vendors it has never seen? Happy to write it in whatever shape the models you already support use.
Scrapegraph-ai carries its own table of models and their context windows, and the open issues show how hard that is to keep honest: a model name that is not in the table falls back to a small window and truncates the page without saying so, the effective window is never reported back to the caller, and the table keeps needing corrections as vendors move. A scrape also burns a lot of tokens in one run, so which model sits behind it becomes a cost question before it is a quality one.
BlockRun serves chat, image, video, music, speech and sound effects behind one OpenAI-compatible base URL, and its model list can be read at runtime rather than assumed, with each call settling in USDC over x402 instead of against an API key. https://blockrun.ai
Would a provider like that be worth carrying, or does the model table already cope with vendors it has never seen? Happy to write it in whatever shape the models you already support use.