GPT-5.4 - Out Now!

Announcement · Pricing · Cursor Docs


GPT 5.4 is now available in Cursor!

We’ve found it to be more natural and assertive than previous models. It’s currently the leader on our internal benchmarks.

10 Likes

for shuuuure

wonderful, but MAX-mode only?

7 Likes

Yeah that’s what mine is showing too, which threw me off. Curious if intentional.

@Colin Wasn’t 5.4 supposed to support 1M token context window? Currently Cursor is only showing 200k

1 Like

Good point, docs also show only 272k for max mode, is there plans for 1M?

The model seems intelligent and fast, only downside is that it spends its time commenting the state of compilation in rust using expensive output tokens instead of just waiting it completes.

Why are the GPT-5.4 models all in MAX mode by default?

They have the same context window as 5.3 and are supposedly more token-efficient. I really don’t understand this choice.

4 Likes

Hey there!

Starting with GPT 5.4, if you’re on Legacy, Request-Based Pricing (as opposed to usage-based pricing billed by token), accessing some frontier models will require Max Mode.

This will not enable the 1M-token context window (or its higher price), but it does mean you’ll be charged at API rates when using GPT 5.4 rather than consuming a fixed number of requests. A separate model identifier (gpt-5.4-1m) will soon be available if you want to use that larger context window.

2 Likes

Damn, that’s a bummer. I’m going to augment as they have 5.4 for free within their IDE

Thank you!

Correct me if I’m wrong but for legacy customers. Do we have to wait until next month to be able to use the new model without MAX mode?

1 Like

If all future flagship models are Max Only, then it’s time to unsubscribe from Cursor :downcast_face_with_sweat:

17 Likes

well i agree as well

1 Like

I understand 1M context window is “Max Only”, but wtf is that 272K context window is “Max Only”…

7 Likes

You say “I charge you with API-based pricing”.

Then why not carry over unused API requests to next month?

It is unfair…

7 Likes

This one is the killer!

So, Cursor is still technically providing a request-based system, but they’re completely changing the definition of a request, as instead of one request taking up ONE REQUEST. Now, a single request can easily use 100 requests.

This isn’t at all a surprise. Cursor has repeatedly engaged in this borderline illegal, shady behaviour, constantly going back on their word and changing pricing structures and features people have already paid for.

9 Likes

I don’t think this is a good approach.

From a user perspective, this feels less like a product improvement and more like a way to push Legacy users from request-based pricing into the newer usage-based model. For long-term Legacy users, it feels inconsistent and unfair.

I think Cursor should consider a fairer and more consistent pricing model for Legacy users, especially one that keeps the fixed-request structure they originally signed up for.

Legacy users are typically some of Cursor’s earliest supporters. I’ve been using Cursor long enough to almost witness its transition from the old tab-based experience to the current agent-based workflow. Even when there were bugs, I was still willing to report them on the forum and wait for fixes. A more thoughtful and stable pricing policy is expected, rather than being gradually optimized away and punish the earliest users.

Cursor’s revenue growth should come from a broader product strategy, not from making the experience less predictable or less favorable for the users who supported the platform early on.

8 Likes

Classic cursor bait and switch. They haven’t learned anything.

7 Likes