New Claude models that just dropped

Discussions
1

I understand there are several threads on this, and of course there is documentation. But is it possible to have a summary of recommendations in terms when to use which model or typical use case scenarios? The tooltip is not adequate and in fact is a bit confusing as it says the same thing, pretty much, for each one of them (Claude 4.6 thinking/ max/ fast and the various combinations) …
Secondly, I only realized that all of these models were enabled by default in the settings (not sure what that means and if there is some kind of context based auto-switching) when I maxed out on token usage for my Pro plan. It all happened very quickly and in the middle of a debugging session, so I had to spend money for ‘on demand’ usage. Not ideal, of course, and would like to know how such situations can be avoided in the future following installation of updates etc.

1 Like
3

Model Recommendations:

Heavy / Ambiguous Tasks

  • GPT 5.3 Codex High

  • Opus 4.6

Small / Medium Tasks

  • Composer 1.5

  • GPT 5.3 Medium

  • Composer 1

  • Gemini Flash 3

All models are enabled by default for every provider to make it easier for users to choose. If you don’t use a model, it won’t be included in your billing.

If you feel your tokens are burning too fast, try to:

  • Create a new chat for each feature

  • Turn off MCPs that you don’t use

For me, I avoid complex tasks at the beginning. I break them down into small or medium tasks first. That way, I can rely on the cheaper models.

Hope this helps you.

5

Thanks a lot, this is helpful. Still wondering about the Claude model distinctions (4.6 and its variations) since I prefer Claude over GPT based on my coding experience. Also, examples of heavy vs. light (number of lines of code, application types) would be good to have.

6

This is a rapidly changing field - what’s valid today won’t be tomorrow. Because there is no said list of what’s good whats bad - this has been stressing quite many people and I think you will find this relatable: Model selection + usage limits are becoming stressful - #11 by valentinoPereira

7

Thanks! I looked at the post you mentioned and agree with it for the most part. I would love for Auto to do the right thing, but it raises the same question/ concern noted in this post (which may or may not lead to a higher pricing as such):

https://forum.cursor.com/t/auto-seems-to-select-heavy-thinking-models-regularly/151876/4

Regarding Ask vs. Agent, I do wish even that was more automated. The prompts which are very clearly a question still very often result in actual implementation/ code changes in Agent mode which can be a bit exasperating. But it is a separate peeve. :slight_smile:

-Sravasti

1 Like