The automatic model selection means v0 can use a different model for reasoning than it uses for planning and again a different model for actually generating the code. For example it might pick GPT 5 for planning or design tasks and then switch to v0-1.5-md to reliably output the code
For this reason the model selector and the GPT-5 switch have been removed
Watch it in action:
I’m making this thread to consolidate early feedback on the feature so the v0 team can iterate on it quickly
Please be constructive in the comments
if you are hitting bugs please provide a chat ID and supporting evidence so we can try to reproduce it
if agent mode as a workflow doesn’t feel suitable for you, please explain more about your use-cases and reasoning so it can be taken into account for future planning iterations
I’ve noticed that my V0 workspace has been migrated from v0.dev to v0.app and now defaults to the new Agentic mode. While I understand the intent to streamline the experience, this change has removed the ability to manually choose the model (e.g., GPT-4o, GPT-5o) for my prompts.
For my workflow, manual model control is critical. I often need to:
Compare outputs from different models for the same prompt.
Ensure deterministic behavior in production prototypes.
Optimize for cost or speed depending on the task.
With Agentic mode always enabled, I have less control over output style and model capabilities, which impacts my ability to deliver precise results for clients and maintain consistency across revisions.
Request:
Please provide an option to disable Agentic mode in my account or for specific projects, restoring the previous manual model selection capability. If a global toggle isn’t possible, even a hidden or developer setting would be extremely helpful.
I appreciate the work going into improving V0 and I understand the vision for Agentic workflows, but for advanced users like myself, the loss of this control is a significant regression.
Looking forward to your guidance on whether this is possible or if a workaround exists.
I’ve tried multiple times to use GPT-5, but the Agent mode doesn’t seem to be utilizing GPT-5 appropriately.
It feels like they’re testing on paying users, which is very uncomfortable. I can tolerate frequent UI changes, but since this is a development tool, changes in the development workflow are extremely confusing.
The agent seems to take a lot more liberties with my prompts as I try to iterate on a design. So it seems like it doesn’t do anything that I’m actually asking. You can take a look at my project for troubleshooting.
If possible, can you revert this agent for iteration?
I do like what it’s doing. It does have a good design taste.
I really like v0 and I don’t want to switch tools, but the agent mode is really taking away my control over the budget and the projects. I’d like to have the option to switch it off or exit it.
Thanks for shipping fast—Agent mode is an interesting direction. However, after the migration to v0.app, I’m seeing a significant drop in output quality for my workflow: the agent’s results are noticeably weaker in visual design quality and content depth, and it takes too many liberties with my prompts. This makes iterative design harder and less predictable.
For my day-to-day, I need tighter control and reproducibility.
Restore manual model selection so I can optimize for cost/speed and compare outputs across models.
Agentic workflows can be valuable, but removing the non-agent path is a regression for advanced users who need precision and consistency. I’m happy to provide example chats or project IDs to help you reproduce the issues.
I haven’t tried it out, but self-picking seems better because I’d definitely much rather have consistently successfully executed prompts at the expense of a few more tokens, but unsuccessful prompts chew tokens anyway so it’s sounds like a no brainer. I’ll try it out later today
Guys, the agent isn’t working well. I feel like I’m using GPT-3.5 because it makes silly mistakes on very simple requests. I hope you’ll switch back to manually selecting the LLM ASAP.
To the V0/Vercel team, I hope this finds you well. Before the migration, I could choose which model to run, such as GPT-4o or GPT-5o. This allowed me to compare outputs for the same prompt, ensure deterministic behavior for prototypes, and make tradeoffs between cost and speed. With Agent Mode always enabled, I no longer have that control. The system decides for me, which introduces variability in style, fidelity, and capabilities from run to run. That unpredictability slows iteration, makes regression testing more difficult, and increases the risk of introducing inconsistencies into a live design.
Since the change, I have also seen a drop in quality for my design workflows. Visual outputs feel less polished, content depth is reduced, and the agent often diverges from the specific instructions in my prompts. This forces me to spend time reworking outputs rather than refining them, which breaks the momentum that is critical in iterative design.
My use case requires stability and control. I need to lock a specific model for the duration of a project, maintain a consistent output style across revisions, and ensure the tool does not alter core behaviors without my direction. If Agent Mode could be made optional, even at the project level, it would allow me to continue working in V0 with the confidence that my design process and delivery quality will not be compromised.
I’ve personally had pretty good luck with agents across multiple projects. It doesn’t feel worse, it still feels fine doing incremental work. You can still confine behavior you don’t want it to do via prompt. I feel like it has just cut down on errors a bit by reviewing it’s own work (Although it’s not close to perfect yet) Just giving my 2 cents. However, I did notice once agents switched, projects that worked just fine completely broke. In my case, it was just a tiny change in the middleware.ts script located inside /lib/supabase.
Just throwing this in here as a bonus, although it might not be appropriate since it’s not related to agents, but dragging a .rar file doesn’t properly port a project over properly anymore. (For some weeks now) It loses some directory/file structure integrity. However, oddly enough, if you re-zip it as a .zip instead of a .rar, it preserves the directory integrity and properly ports over projects.
Shipping this huge change to paying users and taking the money that we spent and charging us for the prompts which i’ve noticed cost more now was a very unwise decision. The Agent is making alot of mistakes and so far i’ve brunt through the gifted credits in less than 2 hours and now it will do the same to the money i paid. add an option for us to revert back and disable agent mode. thank you.
Before the update, everything I asked for, it was able to do. Now, with the update, I’m not getting the same results.
Besides that, it keeps repeating the process over and over, taking a long time to deliver the result — something that used to be quick.
A simple change in my code took more than 10 minutes, when before it took only 2 to 3 minutes, not to mention how many times it repeated the process, just wasting my credits.
On the very first run, it went into a loop on a simple task that I used to complete without any trouble, and it almost maxed out my tokens — this is unacceptable.
Also, they are forcing the use of agents, and I want to be able to disable that.
I’m hating this update, and I think I want my money back.
Hi, I mean this is okay but having the option to choose between models its better like leave us some Control... I dont like the agent mode… choosing between models was better and even more helpful, models were able to generate over 3k code in only 1 file, now this one cant even write 1.5k code, and it laggs a lot.
Agent Mode is very convenient. It’s convenient, but I want to be able to choose the model it runs on. I can only use md all the time. I’d like to be able to use higher-quality models like lg or GPT-5. That’s because md might destroy what I’ve created or misunderstand my intent. I’m willing to accept some credit consumption, so I’d like it reverted.
