[BUG] REGRESSION: Cannot switch between Opus and Sonnet mid-thread in Cowork after April 15 desktop redesign
Status Open
Maintainer reply None cached
Activity 6 comments · opened Apr 16, 2026
Description
After the April 15, 2026 desktop app redesign, mid-thread model switching is broken in Cowork. Previously, users could freely switch between Opus and Sonnet within the same Cowork thread — including within Projects. After relaunching the app post-update, this capability is gone.
This may be related to the regression in #40095 (automatic model switching from Opus to Sonnet), but the behavior here is different: the option to manually switch models mid-thread appears to be entirely removed, not just silently failing.
Environment
- Product: Claude Desktop (Cowork mode)
- Platform: macOS
- Subscription: Max plan
- App version: Post-April 15, 2026 update (desktop redesign)
Steps to Reproduce
- Open a Cowork session
- Start a conversation on one model (e.g., Opus)
- Attempt to switch to a different model (e.g., Sonnet) mid-thread
- Model switching is no longer available mid-thread
Expected Behavior
Ability to switch between Opus and Sonnet mid-thread in Cowork, as was possible before the April 15 update.
Actual Behavior
Model cannot be changed mid-thread in Cowork after the update.
Notes
- This worked reliably in previous versions of the desktop app
- Mid-thread switching was never available in Chat mode (expected), but was available in Cowork
- Likely a regression introduced by the April 14-15 desktop redesign
- Related: #40095
6 Comments
This is not just macOS btw, this is also an issue on Windows version. Also, this is a particularly bad bug given how most people use Cowork. Many use one model to plan and another to execute pieces of given tasks to maximize usage limits and quality of results. This should be a high priority fix.
This is a really bad regression that significantly lowers the usefulness of Claude Cowork. Switching between models during a session helps to make efficient use of the token budget - probably something that Anthropic would like to see as well as it lowers the load on their inference infrastructure. It makes sense to use opus for planning and sonnet for execution - just as I am doing in Claude Code. Being stuck on Opus just unnecessarily burns tokens and lowers customer experience / customer satisfaction.
what's going on with this?, it really hurts in token consumption, i need to plan with opus and execute with sonnet or haiku in same session so not to loose context. please fix it.
+1, and this just bit me hard.
I was deep into an active Cowork task running on Fable 5 when the model became unavailable. Because Cowork locks the model after the first message, I couldn't switch to another available model to keep going, the whole task is now stuck mid-execution, with all its accumulated context, until Fable 5 comes back (whenever that is).
This is a real problem, not a nice-to-have. A model going temporarily unavailable shouldn't be able to freeze an entire long-running task. Every other Anthropic surface, web chat, Claude Code CLI (/model), lets me change the model mid-conversation. Cowork being the only one that doesn't is exactly where it hurts the most, because Cowork tasks are the long-lived ones with the most context to lose.
Please let us change the model at any point in a Cowork task, taking effect from the next message. As it stands, we have zero freedom here, and one upstream availability blip can kill a task we've invested hours into.
This is very frustrating that Claude Engineers cannot release a fix for such a simple issue.
any news on fixing this one?thx