V0 suggestions sometimes spit out junk and counting toward usage fees

v0
1

Hi Team, I have read through forum and I see lots of similar issues are already reported on forum. These repetitive, meaningless tokens still count toward thousands of “consumed” AI tokens

It will be useful to see from team is :

  1. Quality‐gate before counting usage : Only charge tokens for suggestions that pass a minimal “sanity check” (e.g. no more than X repeated substrings, or valid TypeScript syntax).
  2. Usage only upon acceptance : Defer billing until the user actually applies or approves an AI suggestion in their code.

image
image1896×848 207 KB

Also another observation is that :
When I start a project using the md model and later switch to sm for a few queries, my completions degenerate and sometimes outright break existing code.

Why it matters

I expect a smooth fallback—if I downgrade temporarily, it shouldn’t orphan all my previous context. Switching models mid-project can cost time chasing phantom bugs.

Without a warning or compatibility note, I assumed “all v0 models share the same context” and was surprised by the breakage.

Suggested improvements

Model-switch warning: Prompt the user “Switching models will reset your existing context—continue?”

Cross-model context migration: Automatically re-tokenize/re-embed your project history so switching is seamless.

Documentation: Clearly call out in the README or dashboard how sm|md|lg differ in tokenization and context handling.

Note : I used AI to prepare this and share as constructive feedback

4

Context is not lost when you switch models, but a few factors can be at play here

  • the messages generated by one model may not match the same token patterns as another, and therefore aren’t re-interpreted in the same way
  • prompt caching reuses the message history, so the first prompt after switching back to a model may require the model to parse all of the messages that you had sent to the other ones, degrading its output for that message. I’m not 100% sure if v0 works this way but many systems do

Regarding selectively charging for tokens: we switched to usage based pricing to align our costs with yours. Every one of those repeated tokens costs us money. We implement safeguards to reduce the odds of things like this happening, but they’re a fact of life with LLMs as they currently exist

5

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.