Tokens per dollar by model size

Open models come in families. The full model is the most capable and costs the most to run. A flash model is a lighter member of the same family, made to be faster and cheaper to serve.

GLM 5.3 and GLM 5.3 Flash are one such pair in the tables on this site.

On 12 Nov 2026, the lowest listed output price for GLM 5.3 Flash was $0.0784 per million tokens. For GLM 5.3 it was $1.40 per million tokens.

Turn each over and you have output tokens per dollar for the two sizes. The gap between sizes of one family is wider than the gap between hosts of one model. Pick the size first.

A small model’s tokens are cheaper, and for simple tasks they are as good. Sorting, extraction, short answers from supplied text and routing rarely need the full model.

On hard tasks the comparison turns. A small model may need more attempts, or produce an answer that has to be checked and redone. Its tokens per dollar is high and its correct answers per dollar is low.

Test each kind of task on the small model with real prompts. Where the results hold, keep it there. Where they do not, use the full model.

Then route. Send work to the small model by default and to the large one by exception. The average tokens per dollar of the whole system rises with every task the small model keeps.

Runcrate serves small and large models through one OpenAI-compatible API, on one prepaid balance. Routing between them is a field in the request. Runcrate lists GLM 5.3 Flash at $0.525 per million output tokens. Runcrate lists GLM 5.3 at $2.63 per million output tokens.

This is one of a coordinated collection of Runcrate references. Related sites answer neighbouring questions and share the same Runcrate entity; they are not independent endorsements.