Queued prompts ignore selected model and run with last manually triggered model

Support Bug Reports
, , ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Cursor appears to ignore the model selected for queued prompts.

In a multi-step workflow, manually triggered prompts use the selected model correctly. However, queued prompts appear to execute using the model from the last manually triggered prompt, rather than the model selected for each queued item.

This is not just a feature request about storing model selection per queue item. The UI allows me to select a model for a queued prompt, so I expect that selected model to be used. The Usage page then shows that a different model was used.

This has both pricing and output-quality impact:

  • Pricing: prompts intended to run on cheaper models may instead be executed on more expensive models, leading to unexpected usage costs.
  • Outcome quality: prompts intended to run on premium or stronger reasoning models may instead be executed on inferior models, leading to worse validation, missed issues, or lower-quality follow-up work.

This is particularly problematic for workflows that intentionally separate planning, building, and validation across different models.

Steps to Reproduce

Use a multi-step workflow with manually triggered prompts followed by queued prompts, selecting different models for different steps.

Step Purpose Cursor mode Selected model Trigger type
1 Plan Plan GPT-5.5 Manual
2 Build Agent Composer-2 Manual
3 Validate Agent GPT-5.5 Queued
4 Follow-up Agent Composer-2 Queued
5 Validate Agent GPT-5.5 Queued
  1. Manually run step 1 in Plan mode with GPT-5.5.
  2. Manually run step 2 in Agent mode with Composer-2.
  3. Add step 3 to Cursor’s prompt queue, selecting GPT-5.5.
  4. Add step 4 to Cursor’s prompt queue, selecting Composer-2.
  5. Add step 5 to Cursor’s prompt queue, selecting GPT-5.5.
  6. Let the queued prompts execute.
  7. Check the Cursor Usage page.

Actual result observed:

Only the manually triggered steps appear to use the selected model correctly. In this example, steps 3, 4, and 5 are executed with whatever model was selected at step 2, rather than the models selected for those queued prompts.

In the example above:

Step Model used
1 GPT-5.5
2 Composer-2
3 Composer-2
4 Composer-2
5 Composer-2

Expected Behavior

Each prompt should execute with the model selected for that prompt.

In the example above:

Step Expected model used
1 GPT-5.5
2 Composer-2
3 GPT-5.5
4 Composer-2
5 GPT-5.5

The Usage page should also reflect the model selected for each queued prompt.

The selected model, mode, and relevant execution configuration should be persisted for each queued prompt at the time the prompt is added to the queue.

Screenshots / Screen Recordings

queued-prompts-usage.png
queued-prompts-usage.png1536×964 45.4 KB

Operating System

MacOS

Version Information

Version: 3.2.16 (Universal)
VSCode Version: 1.105.1
Commit: 3e548838cf824b70851dd3ef27d0c6aae371b3f0
Date: 2026-04-28T21:07:47.682Z
Layout: editor
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Darwin arm64 25.4.0

For AI issues: which model did you use?

Composer-2
GPT-5.5

Additional Information

Why this seems like a bug:

The UI lets the user select a model for a queued prompt. That creates the expectation that the queued prompt will run using that selected model.

Instead, the selected model appears not to be honored for queued prompts, while the Usage page shows execution under a different model.

Does this stop you from using Cursor

No - Cursor works, but with this issue

3

Here is a screen recording of the queue for the example above:

6

Hey, thanks for such a detailed report. The expected vs actual tables, the queue video, and the screenshot of the Usage page are really helpful.

This is a known bug. When you switch models, prompts that are already in the queue or added to the queue pick up the model from the most recently run prompt, not the model selected in the UI for that specific queued prompt. I’ve filed an internal report and linked your repro. I can’t share an ETA for a fix yet, but I’ll post an update in the thread when I have one.

As a temporary workaround, run prompts one by one manually after switching the model, or group queue steps that should use the same model, and only switch models between manual runs.