How to maximize Opus 4.6 usage?

I currently pay for Ultra and usually on average it’s around $500 on top of Ultra price. Can I get more usage for Opus / Sonnnet if I instead / in addition pay for Claude Premium 20x and use API key with Cursor?

What’s best strategy to get max Opus for the money?

Ok, I’ve learned that API key is not part of Max 20x subscription so it’s impossible to get more usage in Cursor using Claude Max subscription.

1 Like

hi @pharod have a look which tasks you can offload to Auto / Composer 1.5 as we offer there a good sized pool separate from API usage. That way you preserve your Opus usage for the most complex tasks.

Any regular tasks, discovery or trying things out can be handled well with Composer 1.5

Check if you do any of following and if that may help:

  • Plan mode reduces follow up requests
  • Debug mode helps with very focused debugging, saving tokens and better issue resolution.
  • Avoid manually attaching files, logs, rules etc.. as Agent can discover them very efficiently.
  • Check which of your Rules may be omitted as models are getting better, only add them if you notice the Agent making same mistake a few times.
  • Keep chats short and focused on single task.

I’ve tried to use Auto but results were not great and it felt like I have to spend 2-3x more time curating and reviewing and fixing. My usual workflow - design relatively big feature with Plan mode + input from GPT or Claude , build it with Opus. In rare occasions Opus is stuck and I switch to Codex 5.3

One thing I wanted to comment though - in my experience it can’t actually access logs etc. it can’t even access console logs or network requests from own browser tab. I have to manually copy paste from devtools to agent literally always.

Debug mode is amazing, I wish cursor had this clever trick with redirecting console logs to some text file it actually can read by default. I fixed some crazy bugs with debug mode, I really like it.

1 Like

Thanks for the feedback, when did you try Auto last? It has greatly improved since Jan.

Internal browser should be accessible. We saw a few reports where agent couldn’t find the tools but overall it works.

Which Cursor version are you on?

I do this in Opencode to keep Opus costs down, I am sure it’s possible in Cursor. What you need to do is start taking more advantage of sub-agents

e.g. it’s a waste of tokens/money to get Opus to do code searches, call basic MCP like context7/figma, etc. You’ll need to add more rules/etc into cursor to force it to rely more on subagents as by default it is rather conservative

You really want Composer 1.5 to be doing those basic tasks, since the cost comes out of auto usage.

Test building your setup using a cheaper model of course! (Haiku or Sonnet)

In Opencode, I use Kimi to “build” plans from Opus. That’s worth a try or using Composer instead for this.

Another tip would be to use Codex for reviewing code if you use Opus for that