# GPT-5.4 - Out Now! **URL:** \ **Category:** Release Discussions\ **Tags:** openai\ **Created:** [March 5, 2026, 6:20pm UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744 "2026-03-05T18:20:47Z")\ **Posts on this page:** 20\ **Page:** 3
**Author:** ![JimKarter](https://avatars.discourse-cdn.com/v4/letter/j/b3f665/32.png) [@JimKarter](https://forum.cursor.com/u/JimKarter)\ **Post date:** [March 8, 2026, 12:58pm UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/57 "2026-03-08T12:58:25Z")
Anyway Cursor are forcing us to migrate Legacy Request-Based Pricing to API-Based Token Billing. Without our acknowledgements. Wow. ---
**Author:** ![qiuyedx](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/qiuyedx/32/100584_2.png) [@qiuyedx](https://forum.cursor.com/u/qiuyedx)\ **Post date:** [March 9, 2026, 2:34am UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/60 "2026-03-09T02:34:54Z")
Just before this change, I just renewed my cursor subscription for a year! Extremely annoying! How do I cancel the subscription and get a refund? ---
**Author:** ![checkingin](https://avatars.discourse-cdn.com/v4/letter/c/ac8455/32.png) [@checkingin](https://forum.cursor.com/u/checkingin)\ **Post date:** [March 10, 2026, 12:00pm UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/61 "2026-03-10T12:00:45Z")
I should have learnt my lesson from their prior bait and switch tactics. I literally just bought an annual sub a few days ago. ---
**Author:** ![Zr93](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/zr93/32/10997_2.png) [@Zr93](https://forum.cursor.com/u/Zr93)\ **Post date:** [March 10, 2026, 5:01pm UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/64 "2026-03-10T17:01:18Z")
![image](https://us1.discourse-cdn.com/cursor1/original/3X/f/c/fc3161f9fa8ee9724d80c61f776f6fefe1eda368.jpeg) This is information I obtained from X, which means that this really seems to apply only to legacy users. I would like to ask whether all future GPT-series models and Claude-series models will follow this same approach? This kind of differential pricing is deeply problematic. Even if you intend to change the pricing model, you should at the very least make an official announcement and make the billing mechanics transparent, clearly stating how many tokens a single request equal to, rather than switch them quietly until we ask. Alternatively, reverting to a fixed request cap would be far more reasonable. ---
**Author:** ![dylan\_wang](https://avatars.discourse-cdn.com/v4/letter/d/4491bb/32.png) [@dylan\_wang](https://forum.cursor.com/u/dylan_wang)\ **Post date:** [March 11, 2026, 12:17am UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/68 "2026-03-11T00:17:22Z")
email from cursor: Regarding the subagent model: I can see from your screenshot that you configured your code-review subagent with `model: GPT-5.2`, which should cost only 1 request. Your frustration here is completely understandable — you did the right thing by specifying a cheaper model. Unfortunately, on the request-based plan (500 requests/month), **subagents currently default to the Composer model regardless of what model you set in the YAML configuration**. This is a **known limitation** that the team is actively working on improving. To get your subagent model configuration respected, you **would need to enable Max Mode** , which **switches to token-based pricing**. Whether that’s more cost-effective depends on your usage patterns. 🙂🙂🙂🙂🙂🙂🙂🙂🙂 ---
**Author:** ![coolboy5398](https://avatars.discourse-cdn.com/v4/letter/c/a3d4f5/32.png) [@coolboy5398](https://forum.cursor.com/u/coolboy5398)\ **Post date:** [March 11, 2026, 2:27am UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/69 "2026-03-11T02:27:44Z")
I think Cursor is planning to drive away all its longtime users. ---
**Author:** ![StubFeed](https://avatars.discourse-cdn.com/v4/letter/s/41988e/32.png) [@StubFeed](https://forum.cursor.com/u/StubFeed)\ **Post date:** [March 11, 2026, 4:40am UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/71 "2026-03-11T04:40:09Z")
when codex 5.4 ? codex 5.3 extra high vs gpt 5.4 (not codex) = codex still is better in my opinion ---
**Author:** ![NightStars](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/nightstars/32/2666_2.png) [@NightStars](https://forum.cursor.com/u/NightStars)\ **Post date:** [March 11, 2026, 9:49am UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/72 "2026-03-11T09:49:43Z")
Why out of every model, GPT 5.4 was chosen to be MAX-only for legacy pricing? If it were to be 1M mode of the model, that would make sense, but even 270K? Unfair, that is so unfair, Opus 4.6 is allowed to be not MAX-only but all of GPT 5.4 are MAX-only? Makes us want to move sadly considering the features that no other product have (Docs, a proper and fast web search tool, fast tab) ---
**Author:** ![dylan\_wang](https://avatars.discourse-cdn.com/v4/letter/d/4491bb/32.png) [@dylan\_wang](https://forum.cursor.com/u/dylan_wang)\ **Post date:** [March 11, 2026, 11:18am UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/75 "2026-03-11T11:18:52Z")
The next opus model will definitely be the max model, because it’s designed to force legacy users to relinquish their rights. We will see. ---
**Author:** ![dylan\_wang](https://avatars.discourse-cdn.com/v4/letter/d/4491bb/32.png) [@dylan\_wang](https://forum.cursor.com/u/dylan_wang)\ **Post date:** [March 11, 2026, 11:22am UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/77 "2026-03-11T11:22:37Z")
Furthermore, the current flagship model will gradually be marked as deprecated, just like Sonnet 3.5, forcing legacy users to switch entirely to token-based billing. ---
**Author:** ![valentinoPereira](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/valentinopereira/32/78806_2.png) [@valentinoPereira](https://forum.cursor.com/u/valentinoPereira)\ **Post date:** [March 11, 2026, 2:28pm UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/79 "2026-03-11T14:28:37Z")
No my friend, all opus, sonnet and codex models will be max mode only now from March 16, 2026 ![Screenshot 2026-03-09 at 15.40.25](https://us1.discourse-cdn.com/cursor1/original/3X/5/b/5b454dd898268c5748fc1c52a8213f3d83590588.png) ---
**Author:** ![Pat1](https://avatars.discourse-cdn.com/v4/letter/p/43a26b/32.png) [@Pat1](https://forum.cursor.com/u/Pat1)\ **Post date:** [March 11, 2026, 3:11pm UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/80 "2026-03-11T15:11:35Z")
Oof, that is with a heavy heart probably going to be my exit. ---
**Author:** ![dylan\_wang](https://avatars.discourse-cdn.com/v4/letter/d/4491bb/32.png) [@dylan\_wang](https://forum.cursor.com/u/dylan_wang)\ **Post date:** [March 11, 2026, 10:10pm UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/82 "2026-03-11T22:10:52Z")
This is illegal, I choose to unsubscribe. ---
**Author:** ![Colin](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/colin/32/90452_2.png) [@Colin](https://forum.cursor.com/u/Colin)\ **Post date:** [March 11, 2026, 11:06pm UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/84 "2026-03-11T23:06:00Z")
To add some context to the e-mail @valentinoPereira shared – this particular change (Opus, Sonnet, and GPT 5.3 Codex requiring Max Mode) **only** affects Team and Enterprise plans who are on legacy, request-based pricing. It does not affect: - Individual plans (on either legacy, request-based pricing or usage-based pricing) - Team and Enterprise plans on usage-based pricing ---
**Author:** ![valentinoPereira](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/valentinopereira/32/78806_2.png) [@valentinoPereira](https://forum.cursor.com/u/valentinoPereira)\ **Post date:** [March 12, 2026, 7:22am UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/85 "2026-03-12T07:22:14Z")
Hey Colin, just letting you know that this move is highly discouraging. It has sparked uncertainity and loss of trust in your company - and we have already begun talks of moving away from your software. Please share this with your management team that this is a bad move for your business. You’re no longer a startup anymore to keep making changes as you please thinking you’ve got only a few individuals using your software - you’ve got corporations with thousands of team members using it. ---
**Author:** ![ofDamon](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/ofdamon/32/100986_2.png) [@ofDamon](https://forum.cursor.com/u/ofDamon)\ **Post date:** [March 12, 2026, 8:00am UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/88 "2026-03-12T08:00:22Z")
If opus 4.7 is released, will request-based users also only be able to use max-only mode? ---
**Author:** ![liquefy](https://avatars.discourse-cdn.com/v4/letter/l/edb3f5/32.png) [@liquefy](https://forum.cursor.com/u/liquefy)\ **Post date:** [March 12, 2026, 10:49am UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/89 "2026-03-12T10:49:01Z")
I honestly think it is highly respectable move from Cursor to continue supporting legacy pricing for individuals. Although I am not part of this group, I must say good move! ---
**Author:** ![AndreasM1](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/andreasm1/32/50651_2.png) [@AndreasM1](https://forum.cursor.com/u/AndreasM1)\ **Post date:** [March 12, 2026, 10:59am UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/90 "2026-03-12T10:59:14Z")
I was afraid of this… especially because some people here immediately drew comparisons to Opus 4.6. That’s a real shame. Then you can basically forget about Cursor. This has made it very easy for a large number of customers to either switch to the new pricing or cancel their subscriptions altogether and look for something else. I think it’s a real shame. You’d think Cursor wouldn’t need to do this… because they make enough profit either way. ---
**Author:** ![AndreasM1](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/andreasm1/32/50651_2.png) [@AndreasM1](https://forum.cursor.com/u/AndreasM1)\ **Post date:** [March 12, 2026, 3:15pm UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/92 "2026-03-12T15:15:07Z")
Unfortunately, communication on the part of Cursor is really terrible. What about those who don’t regularly read this forum? Will they simply be thrown in at the deep end on March 16 and presented with a fait accompli? I would have found it much better if this had been communicated clearly. It doesn’t make it any better, but at least we can plan for the rest of the month. Besides, in my opinion, the change is too sudden. Cursor should discuss this internally again and offer a transition period. It’s totally unprofessional if it really works the way it was described in the screenshot above. Not every legacy user received this message. A few weeks’ lead time would really be necessary… or a month to be able to replan. * * * This not only restricts access to new models, but also to existing ones. I could have lived with it if they had only applied it to new models from now on, as with GPT4.5. But applying it harshly to existing models is unacceptable. In my opinion, that’s not okay. Not without a longer transition period. In my opinion, this is a very customer-unfriendly approach. You really can’t do something like that. No reputable company would handle it this way. Especially since the users who were there from the beginning and acted as beta testers also contributed to the success. ---
**Author:** ![Zr93](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/zr93/32/10997_2.png) [@Zr93](https://forum.cursor.com/u/Zr93)\ **Post date:** [March 12, 2026, 3:18pm UTC](https://forum.cursor.com/t/gpt-5-4-out-now/153744/93 "2026-03-12T15:18:55Z")
Modifying the GPT5.4 series was already problematic, now is completely destroying the experience for long-time users. Don’t you guys care about user feelings at all? Even if it doesn’t affect individual users now, given the current behavior, who can guarantee what will happen in the future? This change is completely unpredictable, and as one of your earliest users, I’m truly frustrated. [Previous page](https://forum.cursor.com/t/gpt-5-4-out-now/153744.md?page=2) [Next page](https://forum.cursor.com/t/gpt-5-4-out-now/153744.md?page=4)