Image and PDF limits by model

Each row is one limit of one model, with the provider page it was read from, the day, and the words the page uses. Image limits come from each provider's vision page, and PDF limits from its PDF page. A limit nobody could read off the provider's own page is marked unknown, with the page that was tried and the day. It is never filled in with a number from memory. This covers every provider whose pages the build could not reach: each of its models has one row for all its image limits and one for all its PDF limits.

The rows marked "Any model, through Salyro" are Salyro's own bounds. They apply on top of whatever the provider allows: a request within a provider's limit can still be refused by Salyro, before it reaches the provider. Salyro's API carries images, not PDFs, so every provider PDF row says so too: it is what the provider's own API takes, not what reaches it through Salyro.

Two of the image limits are not refusals. Over the long edge or the visual-token count, an image is downscaled before the model sees it, and it costs the tokens of the smaller size. The image token calculator shows the size an image is scaled to and what it costs. Image prep for LLMs resizes, converts and compresses an image to the size a model sees, in your browser.

A PDF's page and size limits count the whole request, including everything sent alongside it. Search by model, filter by provider, input or limit, sort any column, and download the rows you see as CSV.