Interestingly, in terms of cost-effectiveness over 30 days, Sonnet was more expensive than Opus. And Opus 4.6 was slightly more cost-effective than Opus 4.5. And both are twice the price of OpenAI’s models.
And I stopped using Sonnet — if I’m already pinned to the wall by bugs and need Claude, then I’m going straight to Opus.
Good point on the cost. Opus 4.6 actually saves money because you don’t have to deal with the constant retries you get with cheaper models. If you’re doing heavy refactoring, Sonnet tends to trip up and waste a ton of tokens on context heavy errors. Even though the Fast variant has a higher unit price, it’s worth it for the time you save not having to debug the model’s mistakes. It’s usually the better deal for big logic tasks.
Surprisingly, Sonnet 4.5 seems stupid to me. Either GPT-5.2/5.3 or Opus 4.6 if you’re willing to spend more money for a different approach to solving the problem (if we talk about complex tasks).
That’s a solid point. Opus 4.6 Fast actually ends up being cheaper because it stays sharp even with massive context windows. Sonnet tends to lose the plot and forces you to re-prompt, which gets expensive. Since Opus usually nails big architectural changes on the first try, you end up using fewer tokens compared to the constant debugging loops you get with cheaper models.
Where did claude-4.6-opus-max-fast and claude-4.6-opus-max-fast-thinkingdisappear to? I can add the former manually and it works but not the latter. What gives?
same thing happened to me so I made this thread. but i can find both of these in model list in cursor settings
Hi, @Colin
Hello, has there been any consideration of introducing an Ultra Plus plan? For example, a $400 subscription that includes $800 in usage credits. The current $400 credit limit is no longer sufficient for my daily work needs. Yesterday, I only implemented a small feature (fewer than 2,000 lines of code). Using 4.6 for planning and 4.5 for implementation alone cost nearly $10.
In my view, Cursor is still the top option when it comes to model orchestration—better than Claude Max. So even though Claude Max is more affordable, I would still prefer to subscribe to the more expensive Cursor plan.
any comment on why this is happening?
wow!that’s soooo nice!
Any good examples to set this up correctly. I have historically not had great luck getting subagents called.
What happens when you try? Do you get an error message?
Fast mode review:
I have used the fast version for an entire day and it is definitely not worth the extra. It is just a tiny bit faster, if at all. Difference between regular Codex and their fast version is much more noticeable than regular Opus vs fast version. Just in case anybody was wondering.
Simply says model not found and does not work.
Also running into this , though, based on the /analytics page the model I was previously using was claude-4.6-opus-max-thinking-fast and when added as a custom model, does not error at all just never appears in the list of available models.
Had same experience in Cursor (just 2 prompts).
In Claude code it seems to make a difference (2..2.5 speed). But still not worth it 
(I read/heard from others)
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.
