Cursor just doubled usage for paying customers, but only inside its own model pool. That's not generosity. It's pricing pressure.
Cursor's newest usage bump looks simple when you see it in the product. More room. Same bill. But the detail you need to read is not the word "doubled," it's the pool Cursor chose to double.
On July 16, Lee Robinson, who works on Developer Education at Cursor, posted on X that the company had "doubled the included usage of Cursor models on all plans," giving users more access to Grok 4.5 and Composer 2.5. A Cursor community forum reply later spelled out the part that matters: the change is permanent, applies to self-serve Pro, Pro Plus, Ultra, and Teams customers on the new pricing, and covers first-party models including Auto, Composer 2.5, and Cursor Grok 4.5.
Look at the catch. Claude and GPT are not included in the doubled pool. Neither is Gemini. Neither is the free Hobby tier. Older enterprise request-based plans are also excluded, according to the same Cursor forum clarification. If you're paying $20 a month for Pro and you mostly use Auto or Composer 2.5, you've just gained more working room before you hit the limit. That's useful. If your team lives inside Claude or GPT through Cursor, the headline change barely touches your actual workflow.
The discount is a model strategy
Cursor also attached a temporary Grok 4.5 sweetener to the broader usage change. The company says Grok 4.5's included usage is doubled through July 21, 2026, and its published pricing is $2 per million input tokens and $6 per million output tokens, with a faster version at $4 and $18. Cursor's July 8 launch post says the model was trained jointly with SpaceXAI as a mixture-of-experts system and is available across desktop, web, iOS, CLI, and its SDK.
That line matters. Cursor is not only reselling other labs' intelligence anymore. It wants you using the models it helped build, inside the editor it controls, with a cost structure it can manage more directly. Cursor says Grok 4.5 solves multistep tasks in under half the steps of comparable frontier models. That's the pitch. The same post also includes a useful caveat: Cursor said an earlier snapshot of its own codebase was accidentally included in training, giving Grok 4.5 an advantage on CursorBench, and that data has been removed for future models.
Fair enough. That kind of unglamorous caveat is exactly the fact you should keep in mind before treating any benchmark chart as gospel.
Cursor has money to make this move. TechCrunch reported in November 2025 that Anysphere, Cursor's parent company, raised $2.3 billion at a $29.3 billion valuation, with Accel, Coatue, Thrive Capital, Nvidia, Google, Andreessen Horowitz and DST Global among the investors in or around the round. Then the story changed again. Reuters reported on June 16, 2026 that SpaceX agreed to buy Anysphere for $60 billion in an all-stock deal. This isn't a company squeezed for cash.
Attention is the squeeze.
The pressure is coming from outside the editor
The timing isn't accidental. Moonshot AI announced Kimi K3 on July 16, the same day Robinson's post about Cursor usage started circulating. Tom's Hardware reported that Kimi K3 is a 2.8 trillion parameter open-weight model and topped LMArena's Frontend Code Arena with a score of 1,679, ahead of Claude Fable 5 at 1,631 and GPT-5.6 Sol at 1,618. The promised weights are due July 27, according to reporting and technical previews published after the launch.
That is the real pressure. A Chinese lab can ship a huge coding model, let developers try it, promise the weights within days, and immediately force American toolmakers to explain why their $20 seat still deserves the renewal. The answer can't just be brand. Developers check results. They check latency. They check the bill.
Cursor is not alone in pulling this lever. Anthropic said on May 6 that it doubled Claude Code's five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans after signing a compute partnership with SpaceX. GitHub said its Copilot plans moved to usage-based billing on June 1, with new flex allotments for Pro and Pro Plus and a Max plan for heavier use. Windsurf also shifted away from its old credit-style limits earlier this year and moved closer to the same $20 developer-tool price band.
So if you're already paying for Cursor, don't upgrade on habit. Open your usage dashboard and look at which pool you're actually burning. A founder using Composer 2.5 all day may get real value from the doubled first-party allowance. A team routing most work through Claude or GPT inside Cursor should treat the change as noise until the third-party pool moves too.
The bigger point is blunt. AI coding tools have run out of easy sticker-price room. Cursor Pro and Windsurf Pro sit around $20 a month. So does Claude Code Pro, while serious users can blow past bundled usage quickly once agents start running multi-step jobs across a codebase. When the monthly price stops moving, the fight shifts to included usage and model ownership. And whether the tool can keep you inside its own cheapest lane.
Cursor's move gives paying users more room today. It also tells you where the company wants the market to go next: fewer expensive third-party calls, more first-party model usage. And a developer who feels the limit later than they did last week.
Also read: SkyPilot Raises $20 Million to Let AI Teams Shop Compute Across Every Cloud • Alibaba Says Qwen3.8-Max Beats Fable 5 on Price, but Skips the Benchmark • London startup Humanoid becomes Europe's first humanoid robotics unicorn
