Why the mix matters

The mix is how many input tokens a workload uses for each output token. A job that sends nine tokens for every one it gets back has a mix of nine to one. A job that gets back as much as it sends has a mix of one to one.

The mix belongs to the workload, not to the model. The same model gives different tokens per dollar to two jobs with different mixes.

Think of a bundle. For a mix of three to one, the bundle is three input tokens and one output token. Price the bundle: three times the price of an input token plus one times the price of an output token.

Divide a dollar by the price of the bundle to get bundles per dollar. Multiply by four, the tokens in each bundle. That is tokens per dollar at your mix. Change the mix and the bundle changes, and so does the answer.

At a mix heavy in input, the answer sits close to input tokens per dollar. At a mix heavy in output, it falls toward output tokens per dollar. The wider the gap between a model’s two prices, the further it moves.

On 12 Nov 2026, the lowest listed input price for GLM 5.3 was $0.07 per million tokens. The lowest listed output price was $1.40 per million tokens. Every workload’s figure for that model lies between the two that those prices give.

Answering from retrieved documents, classifying, extracting and summarizing are input-heavy. Agents are too, because each step resends the steps before it. Writing from a short brief, translating and generating code are near even or output-heavy.

Do not assume. Each API response reports its input and output tokens. Log a few hundred real requests and divide.

Runcrate’s API is OpenAI-compatible, so the client that reads token counts for you today keeps working. One prepaid balance pays for all 125 models. Runcrate lists GLM 5.3 at $0.59 per million input tokens and $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.