---
title: "Models and capabilities"
description: "How RShift routes model calls and what to do when a model is unavailable."
image: "https://help.rshift.com/og.svg"
---

> Documentation Index
> Fetch the complete documentation index at: https://help.rshift.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Models and capabilities

RShift calls a set of replaceable models through shared model collections named `rshift-{capability}-models`. The name never contains a specific model version or provider, so RShift can change the underlying model without changing your workflows.

Capability codes are fixed and semantic: `tts` (text to speech), `stt` (speech to text), `trans` (translation), and `privacy` (privacy-sensitive processing).

## Availability

Model availability depends on your plan and enabled capabilities, AI credit balance and workspace budget limits, provider status, and whether the input fits the capability’s limits.

## If a generation fails

1. Check your plan and AI credits in billing.
2. Retry, or switch to another available model.
3. Shorten the input or split the task.
4. If files are involved, confirm size and format.

See [Failed generations](/rshift-ai/failed-generations/).

Source: https://help.rshift.com/rshift-ai/models-and-capabilities/index.mdx