On July 8, Cursor and SpaceXAI jointly released Grok 4.5 — officially billed as their "most intelligent model yet" and the first Cursor model not trained exclusively for software engineering. During launch week, all subscription tiers received double included usage (effectively 50% off), letting you run the same token volume on half the pool.
The promo was originally set to end July 15. Because some users did not receive the discount correctly during model slug migration and billing attribution got messy, Cursor's community team confirmed an extension to July 21 and issued credits to affected users. If you missed Grok 4.5 last week or saw odd billing — you still have a window to make the most of this deal.
I. TL;DR: Is this deal worth it?
Yes. Even if you're not a heavy Cursor user, running one or two real projects on Grok 4.5 during the double-quota window (refactors, cross-file debugging, long Agent chains) cuts cost nearly in half. Grok 4.5 targets hard, long, multi-tool tasks — Cursor claims it completes multi-step work in fewer than half the steps of comparable frontier models. Same pool, potentially fewer rounds to finish more work.
Deal essentials
- Deadline: Extended to July 21, 2026 — confirm in Dashboard
- Billing pool: First-Party Models (not the API pool)
- All platforms: Desktop, Web, mobile, CLI, and SDK all qualify
II. What is Grok 4.5, and how to choose vs Composer 2.5
Grok 4.5 is a MoE (Mixture-of-Experts) model co-trained by Cursor and SpaceXAI on trillions of tokens of real developer–Agent interactions. Its scope is broader than coding-focused Composer 2.5 — software engineering, data science, finance, law, and other "complex work done on a computer" all appear in the training mix.
Cursor positions them as complementary tiers: Composer 2.5 handles daily fast coding; Grok 4.5 takes hard tasks needing creative tool use, multi-step reasoning, and recovery from errors. Think of Composer as your daily commuter and Grok 4.5 as the off-road rally car — not a replacement, a division of labor.
| Dimension | Grok 4.5 (standard) | Composer 2.5 (standard) | Composer 2.5 Fast |
|---|---|---|---|
| Positioning | Hard tasks, long tasks, cross-domain | Daily coding, fast iteration | Low-latency default |
| Pool consumption | Faster than Composer standard | Baseline | More expensive than Grok 4.5 standard |
| Overage rate | $2/M in · $6/M out | See Cursor pricing page | See Cursor pricing page |
| Double promo | ✓ Extended to July 21 | — | — |
III. Double-quota promo: rules and why it was extended
Timeline
- July 8: Grok 4.5 launches; launch week includes double included usage
- July 8–9: Some requests were incorrectly billed to the API pool (fixed — now unified under First-Party Models)
- July 14–15: Model slugs bulk-renamed (e.g.
grok-4.5-xhigh→cursor-grok-4.5-high); promo did not migrate to new slugs, so some users were charged at list price - July 15–16: Cursor team issued credits; double promo extended to July 21
If Grok 4.5 requests around July 14–15 did not show 50% off, check cursor.com/dashboard/usage for requests marked "free" as compensation. Community members have confirmed receiving credits.
Platforms covered
Double quota is not desktop-only. Cursor confirms Grok 4.5 is available on these entry points, and the promo applies equally:
- Desktop: The familiar IDE — select Grok 4.5 directly in Agent mode
- Web & Mobile: Spawn cloud Agents in the browser or on iOS
- CLI: Run Agents from terminal, scripts, or any editor
- SDK: Build automation pipelines with the Cursor Agent SDK
- Integrations: Slack, GitHub, Linear, JetBrains, and more (depending on plan and config)
Individual plans (Pro / Pro+ / Ultra) and team plans include significant usage. Enterprise uses unified committed usage; Grok 4.5 is still treated as a first-party model, exempt from Cursor Token Rate.
IV. Which pool gets charged? Don't mix them up
This was the most-asked question in launch week. Short version:
Billing at a glance
- Auto mode: Still free on Free plan; paid plans follow Auto logic
- Explicit Grok 4.5 / Composer 2.5: Draws from First-Party Models pool (formerly Auto + Composer pool, renamed July 8)
- GPT / Claude / Gemini, etc.: Draws from API pool
- BYOK + on-demand: API pool or your own key
Note: the First-Party Models pool is consumed at token rates, not "one request = one unit." For the same code volume, standard Grok 4.5 typically depletes the pool faster than standard Composer 2.5 — but compared to Composer 2.5 Fast (the product default), standard Grok 4.5 is actually cheaper. During the double promo, those differences are halved — a good time to try Grok 4.5 High effort.
V. Step-by-step: switch models + verify billing
Desktop switch: In the Agent panel model picker, select "Grok 4.5" and adjust Low / Medium / High effort by task difficulty. Long tasks and cross-repo refactors benefit from High; simple completions can use Low to save quota.
CLI usage: After installing Cursor CLI, specify grok-4.5 model slugs in agent commands (UI display names and backend slugs aligned on July 14 — prefer entries with the cursor-grok-4.5 prefix).
Three-step billing check:
- Open Dashboard → Spending and confirm Grok 4.5 appears under First-Party Models
- Open Dashboard → Usage and check each request's pool attribution and promotional / free markers
- If API pool charges persist, report with request ID on the Cursor forum — the team responds quickly to these issues
VI. 5 real workloads to maximize double quota
The double promo isn't for idle chat — it's for stress-testing the new model on real work. These scenarios offer the best value during the window:
- Cross-file architecture refactors (10+ files): Grok 4.5 uses fewer steps — let it map call chains before editing
- CI/CD scripts and Dockerfile tuning: Multi-tool, multi-round validation plays to its long-task strength
- Adding tests to legacy code: Have it read the full repo and generate tests instead of feeding files one by one
- CLI + cloud Agent combo: Trigger via local Cursor CLI, monitor long runs on Web — quota is shared across platforms
- Pair with Cloud Mac execution nodes: Grok 4.5 plans and edits in Cursor; Xcode / Flutter builds run on a Cloud Mac — model savings help offset execution costs
Pro user tip
If your Pro plan still has First-Party pool headroom this month, use Grok 4.5 High effort on long tasks you'd normally skip during the double window — the same work costs twice as much after the promo ends. Clear your "wanted to try but didn't" Agent experiment list now.
VII. Pitfalls: slug migration, Fast variants, and EU limits
- Slug alignment: After July 14, pick UI entries matching
cursor-grok-4.5-*; old slugs may not receive the promo - Fast variants cost more: Grok 4.5 Fast lists at $4/M input, $18/M output and depletes the pool faster; promo math for Fast may differ from standard — run a small task first and check usage numbers
- EU not yet available: Due to EU AI Act compliance, Grok 4.5 is not open in the EU yet; rollout expected mid-to-late July 2026
- Not unlimited: "Included" means plan allowance, not unlimited calls; after the pool is exhausted, credits or on-demand billing applies
- Pages may lag: The official landing page still says "through July 15" as of publication — extension details are authoritative in Dashboard and forum staff replies
FAQ
When does the double-quota extension end?
Launch week ran through July 15; due to slug migration billing issues, extended to July 21, 2026. Confirm at cursor.com/dashboard.
Which pool gets charged?
First-Party Models pool, shared with Auto and Composer 2.5. Does not consume the API pool.
Which platforms qualify?
Desktop, Web, iOS, CLI, and SDK — individual and team plans covered. EU awaiting compliance rollout.
Which is more quota-efficient vs Composer 2.5?
Depends on scenario: daily quick edits → Composer; hard tasks and long chains → Grok 4.5 with fewer total steps. Grok 4.5 value improves significantly during the double promo.
Can Free plan users use it?
Auto remains free; explicitly selecting Grok 4.5 consumes limited First-Party pool allowance. Heavy users should upgrade.
Conclusion
Cursor Grok 4.5's extended double quota is one of the most notable developer deals of H2 2026. It's not just marketing — Grok 4.5 is a new option for multi-step Agent work, and the double window lets you cheaply test whether handing hard tasks to a frontier model pays off.
Action list: switch to Grok 4.5 in the model picker today and run a cross-file task you've been putting off; open Dashboard and confirm the billing pool; if you have iOS/macOS build needs, offload execution to Cloud Mac so model quota and execution environment each do their job.
Promo ends July 21 — after that, the same Grok 4.5 long tasks return to normal rates. The point of a deal window isn't waiting for a "better model" — it's trying everything you meant to try while the price is right.
Model savings can offset Cloud Mac execution costs
Grok 4.5 plans and edits in Cursor; Xcode builds, TestFlight packaging, and Flutter iOS compilation go to Vuncloud Cloud Mac. Maximize model quota during the double promo, and execution node rental becomes easier to justify.
View Cloud Mac Plans · Why execution nodes are the scarce resource
Related Reading
- Which IDEs Does OpenAI Codex Micro Support? VS Code, Cursor, Claude Code Hands-On Test
- Everything Changes! Codex Fully Merges Into ChatGPT in 2026 — How Scary Is ChatGPT Work?
- The Model Arms Race Is Over—Why Mac Compute Nodes Are Suddenly Hard to Get
- Why Does Cursor Miss Half the Call Sites When Editing Across Files?
Promo rules and deadlines compiled from Cursor's official blog, forum staff replies, and community feedback as of July 16, 2026; subject to change — confirm at cursor.com/dashboard. Grok 4.5 pricing and regional availability per Cursor and SpaceXAI official announcements. Last updated: July 16, 2026.