Claude 4.6 Opus is now available in Cursor!
It’s highly effective at long-running tasks and reviewing code.
Have you tried it yet?
Update: Feb 7, 2026
Opus 4.6 (fast mode) is now available in Cursor as well! It’s 2.5x faster and currently in research preview.
Pricing: $30 per million input tokens / $150 per million output tokens (6x more expensive than regular Opus 4.6).
For the next 10 days, it’s available at 50% off in Cursor!
Thank you! Cant wait to see how it does.
For anybody wondering, as I was:
Maybe a question - when using max mode, are tokens below 200k tokens also priced at max price (double)? or does it apply only to tokens when context exceeds this limit (as anthropic page says)?
Great question @liquefy! It only applies when tokens exceed this limit (and actually dip into the extended context window)
Opus 4.5 really blew my mind, i’m very curious about the capabilities of 4.6! Awesome
Could you explain what’s the difference between Opus 4.6 (Thinking) and Opus 4.6? They both have Version: high effort . Does it mean that they are both thinking versions?
So effective, yet still so expensive 
wen 1million context window
use max mode for 1mil
Adaptive Thinking is enabled for thinking models.
@Colin may you elaborate a little please? if I understand:
non-thinking version: fixed reasoning level to high
thinking version: reasoning is based on request, might think a lot or not at all
is this correct? I am slightly confused.
Effort is set to high, not reasoning/thinking.
That said, as noted in the Anthropic docs:
At
high(default) andmaxeffort, Claude will almost always think.
It’s a bit confusing, I agree! I’ll be sticking with High (Thinking) 
.
Opus 4.6 is amazing Innovation
Thanks
Good timing for the relase, I was definately in need of more reasoning horsepower. Overdue solutions now in progress Thank you Anthropic & Cursor !!
but where sonnet 5?
It’s the best for now.
Interesting experince tonight after a 200 dollar day with Opus 4.6 I noticed a problem surfacing in the running code done by Opus. I made a detailed prompt about the issue, included log files and by mistake submitted the prompt to my “Auto” model thread instead.
The solution by “Auto” on the Opus 4.6 MAJOR bug was sound, and I even got Opus to review it, admitting the corrective work was indeed good.
My bad for not noticing I was using the wrong thread and getting great work done without paying extra.
No one is perfect … yet.
I want to make sure I understand this correctly.
Do I understand correctly that Opus 4.6 (Thinking) uses
thinking: { type: "adaptive" },
while Opus 4.6 omits the thinking parameter entirely?
If that’s the case, what is the reason for displaying “high effort” next to the non‑thinking model as well, the same way it is shown for the thinking version?
Effort is a measure of how many tokens the models spends roughly, and than can be both with or without thinking!
Both selections are high effort! (Thinking) enables adaptive thinking.
