Is Grok 4.5 included in the Free Auto tier? Confused about my usage page

[fywtr]

I’m a bit confused about the usage tracking on my free auto plan.

When Grok 4.5 went online, I saw it was included in the Auto and Composer quota. I tried it out in Cursor, and it showed up as “Free”. However, a few minutes ago, I noticed it started counting against my quota (showing as “Included” with a price tag).

I didn’t receive any notification about this change. What is the actual billing rule for the old free auto plan? Is it only free when “Auto” is selected?

Also, I’ve noticed that the billing and usage pages have changed multiple times without any prior announcement. It really lacks transparency and leaves users in the dark.

Can anyone clear this up? Thanks!

---

Another bug:

Here the Grok recorded as xhigh, But both in Cursor app and usage page, it shows high.

[Duong_Do]

bumping this up.

[KhanhDo (Niteco)]

bumping this up.

[deanrie]

Hey, thanks for the detailed report. I can see the screenshots from the usage page showing Grok 4.5 events and the effort level breakdown. Let me break it down.

Billing on your free Auto plan

Your answer is basically yes. Auto stays free. But when you explicitly pick Grok 4.5, that’s a named first party model, and it uses your included First party models pool. This was renamed from Auto + Composer on July 8. That’s why the usage page shows it as Included plus a per token price. This is expected.

The quick Free label you saw at the start was a launch day glitch. When Grok 4.5 launched, usage was incorrectly attributed to the API pool. This is fixed now, and usage is correctly pulled from the First party models pool. More details in this thread: Grok 4.5 pricing for subscription plans

Small bonus: included Grok 4.5 usage is currently doubled. This is 50% off, but the promo ends on July 15, so there are only a couple days left.

xhigh vs high

This is a known display mismatch in the effort level labels between the model picker app and the usage page. It’s visual only and doesn’t affect billing. Charges are applied correctly. I don’t have an ETA for a fix yet, but the issue is being tracked.

On billing page transparency, feedback noted and I’ll pass it to the team. If anything on the usage page looks off, send the Request ID right from the chat menu using Chat menu > Copy Request ID, and we can check the exact events.

[fywtr]

Thanks for the reply!

The usage page in my current plan does not show the first-party pool and API pool separately. There is only a single “Included Usage” summary section. Because of this, I want to double-check: will using Grok 4.5 and the Composer model count against my API usage?

[deanrie]

Thanks for the extra info. I can see the screenshot from Billing & Invoices, where all items Grok 4.5, Composer 2.5, GPT-5.5 are marked as “Included”, and the total shows $34.60 Included.

Short answer: no, Grok 4.5 and Composer 2.5 do not use up your API quota. Both are first-party models, so they draw from the included First-party models pool, not from API or on-demand. That’s why they show up as “Included” in usage.

About gpt-5.5-medium in the same list: that’s a third-party model, and it draws from your API pool around $20 on Pro, not from first-party. It’s also marked “Included” right now, which means it’s still within your API quota. Keep in mind that once the API pool runs out, GPT-5.5 and other third-party models like Claude or Gemini will be the first to switch to on-demand, not Grok or Composer.

A single “Included Usage” block without splitting by pools is normal for your plan. As long as a line is marked “Included”, it’s within the included limit and won’t count as paid spend. If on-demand kicks in, it would show up as a separate paid line, not under “Included”.

If you see any item that isn’t marked “Included”, send the Request ID via Chat menu > Copy Request ID, and I’ll check the specific events.

[fywtr]

Thanks for your patience and the detailed explanation!

However, my experience from last month felt a bit different. After my API quota ran out, some Composer 2.5 requests actually started drawing from my credits. Unfortunately, I can’t track down the exact Request IDs from last month now.

I’ll keep a close eye on my usage this month. If the same thing happens again, I’ll post it here.

[fywtr]

I have done some simple testing today, and it turns out that Grok 4.5 is actually using my credits now.

Please see the screenshots above. I used two different GPT models, which cost exactly $10 of my API quota total. The remaining requests were from Composer and Grok, which according to the rules, should be included in the first-party pool.

On my billing page, it shows zero usage for the Pro plan, but the usage details show these requests are being drawn from promotions and credits. In fact, I noticed my credit balance decreased. You can see my credits dropped from $221 to $214 compared to the screenshot I posted in post #4. (In my latest test today, it dropped from $215 to $214, costing $1).

Then, I tried using GPT 5.6 Sol in a new session, and it also consumed my credits.

Based on this, it looks like only the $10 GPT model usage should have counted towards the API pool, and I should still have at least $10 of that quota left. Instead, the other models that should be in the first-party model pool are also eating into my credits.

This confirms that both the Composer and Grok models are being incorrectly counted towards my API pool, which matches my suspicion from recent months regarding Composer 2.5 usage.

---

Here are some Request IDs associated with my tests today. Hopefully, they can help you investigate and debug this issue:

Grok 4.5: fc3e9fae-1d3f-4ccd-a69a-9c72b4205128 80a7570e-52a0-4375-8528-244ea860ee95
GPT 5.6 Sol: 2f6ca644-43bb-41bc-b756-70c24fd88ee9 (These session are still in progressing when I submit this post.)

[fywtr]

Also, it seems I can’t switch models in a single session as expected.

In session 2f6ca644-43bb-41bc-b756-70c24fd88ee9, I started with GPT 5.6 Sol to generate a plan, and then manually switched to Grok 4.5 to execute it.

However, even though I selected Grok, the system kept generating GPT 5.6 Sol requests in the background. Because it was too slow, I stopped it and switched to “Auto” — but the logs show GPT 5.6 Sol requests were still being made.

To make matters worse, the billing for these unintended requests is completely inconsistent (as shown in the screenshot):

• One of them was charged as “Included” (using my plan quota).

• But the others (marked as “Free” with the info icon) actually deducted my credits.

This is clearly unreasonable and there must be a bug somewhere.

[deanrie]

Thanks for the detailed testing and for the Request IDs. That’s exactly what we need to check the events one by one. I can see your screenshots with Included Usage, the credit deduction breakdown, and the session from post #12 where, after selecting Grok, gpt-5.6-sol-medium requests kept running in the background.

A couple quick notes so we don’t get mixed up:

• The “Free” label with the info icon does not mean “free outside limits”. It means “covered by promos or credits” (the tooltip says usage provided by Cursor, including promotions or credits). So when you see “Free” on those requests, those are the ones that reduce your credit balance.
• That explains the balance drop, but your main question is separate. First-party models (Grok 4.5, Composer) shouldn’t touch credits or API as long as there’s room in the included First-party pool. We need to verify that at the individual event level.

I’ll take your three Request IDs and pass them to the team so they can confirm which pool each one was billed to:

• Grok 4.5: fc3e9fae-1d3f-4ccd-a69a-9c72b4205128, 80a7570e-52a0-4375-8528-244ea860ee95
• GPT 5.6 Sol: 2f6ca644-43bb-41bc-b756-70c24fd88ee9

About post #12 (you selected Grok, but GPT 5.6 Sol kept running in the background, even after switching to Auto), this looks like a separate issue where the model change inside a session isn’t being picked up. To debug that cleanly, can you share your Cursor version and OS? It might be best to move that into a separate thread so we don’t mix it with the pool attribution question.

[mohitjain (Mohit)]

Building on @deanrie’s reply, here’s the piece that ties it together:

Grok 4.5 and Composer 2.5 don’t draw from your API pool (that pool is only for models like GPT, Claude, and Gemini) - they draw from your included First-party models pool, and Auto stays free. The key detail: that First-party pool is generous but not unlimited. Once you use it up in a billing cycle, further Grok or Composer usage is covered by your credit balance before you’d ever reach paid on-demand. That’s the balance drop you’re seeing, and it’s why those rows show “Free” with the info icon - as deanrie said, “Free” here means “covered by credits or promos,” so it does reduce your balance.

If you think any of your requests were billed incorrectly, email [email protected] with those Request IDs, and the team can review them against your account.

On GPT 5.6 Sol continuing after you selected Grok: that’s a separate issue, and it looks like the mid-session model change wasn’t picked up. If you start a new thread with your Cursor version and OS, we’ll dig into that one directly

[fywtr]

Thanks @deanrie and @mohitjain

That doesn’t make sense.

I just downloaded my usage CSV and did some calculations, and the data completely contradicts your explanation. Here are the facts from my current billing cycle:

1. My actual API usage is well under the limit: Excluding Grok and Composer, all other requests marked as “Included” (using GPT 5.5 and 5.6) add up to exactly $11.31. Since my API pool has a $20 limit, I should still have a remaining balance. There is absolutely no reason for any GPT requests to be drawing from my credits at this point. Yet, as my screenshots show, identical GPT Sol requests right next to the “Included” one were billed to my credits.
2. The “generous” First-party pool is way too small: All of my Grok and Composer requests marked as “Included” add up to only $35.74. If this is where my “First-party pool” supposedly ran out, that is a very low ceiling. It is far below the “generous” quota that both the team claims and the community generally experiences.

I will also send an email to [email protected] with my CSV and Request IDs so the team can look into this billing discrepancy.

---

Regarding the second issue: yes, you are correct—the mid-session model change was definitely not picked up. At that time, GPT Sol was being too slow, so I stopped it and switched to both “Auto” and “Grok.” However, after the run finished, the usage page showed that every single request was still using GPT Sol.

Since I have been using the Nightly version for a long time, I can’t give you an exact static version number for the past events, but I can assure you that I am always on the latest build. You should be able to cross-reference the release time with my request logs to find the exact version. For reference, here is my current Cursor version information:

Version: 3.11.25
VS Code Extension API: 1.125.0
Commit: fc2563ec93d793fc275eef734405a4fdf8b47b20
Date: 2026-07-15T01:27:41.754Z
Layout: Agent Window
Build Type: Stable
Release Track: Nightly
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.256
OS: Darwin arm64 25.5.0

---

By the way, the UI transparency has actually gone backward

Aside from the technical issues, I also wanted to bring this up. It’s currently really hard to calculate how much of my usage is actually drawing from my credits.

On the usage webpage, these requests are just labeled as “Free” (even though they are costing me real money). A few months ago, the dashboard used to show the token count and the equivalent dollar amount for these requests, which was super helpful for keeping track of our budget. But that was removed.

Now, because it just says “Free” on the screen, I have to export the CSV and manually calculate the costs myself based on the raw input, output, and cache tokens. This is extremely inconvenient.

It would be so much better and more transparent if you could show the actual cost on the page, or at least label them as “Included in Credits” instead of just “Free.”

[fywtr]

After exchanging a few emails with [email protected], I’m leaving this final update for anyone following this thread. Honestly, the entire resolution process and the chaotic handling of this issue have been deeply frustrating.

Here is a quick summary of this experience for anyone following this thread:

1. Misleading information across all official channels:
   From the initial announcement email and the official docs (which lack clear details for legacy plans), to support AI assistants and even staff replies on this forum—every source reassured me that Grok 4.5 would not draw from my API quota. To make matters worse, the usage dashboard keeps changing randomly. Right now, even my credit balance display has completely disappeared from my account page.

2. Wasted credits due to contradictory info and system bugs:
   Because official confirmations completely contradicted what I was seeing on my bill, I was forced to perform multiple test runs to troubleshoot. Combined with a mid-session model-switching bug (where the system stuck on generating GPT-5.6 despite being switched to Grok/Auto), I ended up wasting a significant amount of personal credits just trying to figure out their billing rules.

3. Moving on to fresh and more intriguing alternatives:
   While Cursor’s engineering is solid, the constant UI shifts and chaotic billing communication make it really hard to rely on right now. There are so many exciting, fresh models and plans popping up on the market lately—like the newly released Kimi K3—that look much more appealing to try out. I’m going to take a break from Cursor until its product management stabilizes, and go explore those fun new options instead.