Reference

Models and credits

The model list and how Modloom Credits are spent.

You choose a model for each request, from fast and cheap to the heaviest available.

The models#

Models marked Paid plans need a Maker or Studio plan. On the free plan they show a padlock in the picker.

Modloom Credits#

AI work is metered in Modloom Credits. There is no unmetered model. One credit is a fixed amount of work on the reference model, and every other model is priced relative to it.

• A model that costs twice as much per token spends twice the credits for the same work.
• The picker shows each model's rate next to its name, for example x1 or x0.05. x1 is the standard rate.
• Both input and output count, and output is weighted more heavily because it costs more.
• Rates follow the provider's current prices and are refreshed daily, so a rate can change over time.

What affects the cost of a run#

• The model you pick.
• How much the agent reads and writes. Large projects and long threads mean more input.
• How many attempts it takes. Builds that fail and get fixed cost more than builds that pass first time.

A few habits keep runs cheap:

• start with a fast model and move up only if the result is not good enough,
• plan first so the build run is shorter,
• keep requests focused on one change.

Limits per run#

Every run has a ceiling on turns, time and cost so a single request cannot run away. If a run reaches it, the run stops and tells you. Split the request into smaller ones.

Long conversations#

When a thread grows past what the model can hold, older parts are summarized automatically. You see a "Conversation compacted" notice. Details from early in a very long thread can be lost, so restate anything important.

When credits run out#

When your weekly credits are used up, runs are blocked until the next reset, and you get an email. See Plans and limits.