DeepSeek models are the value end of the catalog. They produce polished apps with rich workspace data, at the lowest cost band we offer.
DeepSeek models available today
| Model | Good at | Speed | Intelligence | Cost | Availability |
|---|---|---|---|---|---|
| DeepSeek V4 Pro | Frontier reasoning and coding | Fast | High | Low | Every paid plan |
How to pick within DeepSeek
- V4 Pro for reasoning and coding work where you still want the low cost band.
If you are on the Free plan, use the TSK-Instant rung. DeepSeek V4 Pro needs a paid plan.
Set it on an agent or a chat
Open the model menu in the builder chat or on an agent. In the builder chat, open More models and choose from the provider groups. In the agent editor, choose from All Models. The choice applies only to that surface. An AI step in an automation offers TSK-Auto and the rungs only.
If you do not want to pin a model, a rung sets a speed-to-depth trade-off and follows the catalog forward as models change.
How DeepSeek performs
We build the same apps on every family and publish what happened. See the DeepSeek build results for the test-by-test record.
Next steps
- Compare every family in the full catalog.
- Read how Taskade picks a model for Auto and the rungs.
- Open each DeepSeek model from the DeepSeek models page to see its plans and test results.