Models

Models

Overview

Every prompt, agent, and automation runs through a model. TSK-Auto is the default, and it resolves a model on every turn. A manual choice is optional.

This page explains how that routing works and how to override it. For the catalog of models you can pin, see Custom Engine.

   TSK-Auto            TSK rungs              Custom Engine
   ────────            ─────────              ─────────────
   Picks for you       You set the            You pin one
   every turn          trade-off              exact model

   no setup            Instant → Reasoning    frontier models
                       fast ←→ deep           many providers
        │                    │                      │
        └────────────────────┴──────────────────────┘
                             │
                    same workspace, same token
Level You choose Best for
TSK-Auto Nothing. Taskade routes each turn. Almost everything. This is the default.
TSK rungs A speed-to-depth trade-off You know the job is quick, or you know it is hard
Custom Engine One exact model Comparing models, or one model suits the task

TSK-Auto

TSK-Auto is the default on every surface and on every plan. It carries a Recommended pill in the model menu and needs no setup.

Auto picks a model once per turn. It reads the kind of work in the turn and the plan the workspace runs on, then routes it to a model that fits.

Two properties matter more than the specific choice:

  • Auto re-decides every turn. A build turn and a follow-up edit in the same conversation can land on different models.
  • Auto is not pinned to your workspace. Taskade re-points the Auto lane as models ship and retire, so you get the improvement with no rebuild and no migration.

That second property is why these docs do not name the model behind Auto. Any name printed here becomes wrong the next time the lane moves, and the lane moves on a schedule you do not control.

When to override Auto

Override it when you know something about the job that Auto cannot infer from the turn:

You know Use
The job is trivial and you want it now TSK-Instant
The job is a bigger app that needs more care TSK-Thinking
The job is a complex app and worth more credits TSK-Reasoning
You are comparing two models on one prompt Custom Engine

An override applies only to the surface you set it on. Setting a model on one agent changes no other agent.

TSK rungs

A rung sets a trade-off, not a model. You tell Taskade how much thinking the job deserves, and Taskade maps that to a model for you. It is the middle ground between Auto, which decides everything, and Custom Engine, which decides nothing for you.

   TSK-Instant      TSK-Standard     TSK-Thinking     TSK-Reasoning
   ───────────      ────────────     ────────────     ─────────────
   quick edits      everyday apps    bigger apps,     complex apps,
                                     more care        more credits

   ◀───────────────── faster                 deeper ─────────────────▶
Rung What it is for Availability
TSK-Instant Quick edits Every plan, including Free
TSK-Standard Everyday apps Every paid plan
TSK-Thinking Bigger apps that need more care Every paid plan
TSK-Reasoning Complex apps. Uses more credits Business and up

Match the rung to the shape of the work, not to how important the work feels:

  • A one-line change to text that already exists is a TSK-Instant job. A deeper rung returns the same edit more slowly.
  • A normal build step or agent answer is TSK-Standard.
  • A bigger app that needs more care than an everyday build is TSK-Thinking. It uses more credits for each turn than TSK-Standard.
  • Work that produces a complex app or a system rather than an answer is TSK-Reasoning. It uses more credits for each turn than the other rungs.

A rung survives the catalog. When a provider ships a better model at the same trade-off, Taskade re-points the rung and your agent improves with no edit. A pinned model id does not move, so it ages.

The rungs are also the whole model menu of an AI step in an automation. That menu shows TSK-Auto and the TSK rungs under a Taskade credits heading, with no Custom Engine list. A step that already stores a pinned model keeps it. If that model is retired, or no longer runs on Taskade AI credits, the step runs on TSK-Auto.

Bring your own key

To bill a provider account you control, connect it and add a provider step. Taskade supports OpenAI, Anthropic, Google Gemini, Groq, and OpenRouter accounts. The step runs on your key, and every step around it keeps working normally.

On Pro and up, an agent can also run on your own OpenAI, Anthropic or Google Vertex AI key with Bring Your Own Key. See Claude in Taskade for the agent setup.

See Multi-model AI access for the provider-step setup.

Next steps

Model routing1

Model families10