Cursor pricing changed in June 2025: the $20 Pro plan now includes unlimited normal requests, while Max Mode is billed at the model provider's token cost plus a 20% margin. This article explains both mechanisms with the actual request-per-million-token numbers and a worked cost example.
What changed in Cursor pricing in June 2025
Cursor pricing changed in June 2025: the $20 Pro plan now includes unlimited normal requests, unlimited Tab completions, and background agent access, replacing the old quota of 500 requests per month. Max Mode remains a separate, metered cost. The details below reflect the pricing described by Cursor at that date and may have changed since; always check the current pricing page before relying on any figure.
Under the previous model, each message counted as one request regardless of token volume, and the Pro plan included 500 requests per month. Beyond the quota you could either pay per request or accept slower responses from a low-priority pool. The speaker in the source video, published on June 18, 2025, notes that his own account still displayed the 500-request counter the morning of recording and had migrated to the new plan by the time he recorded.
The new Pro plan markets unlimited calls, but Cursor applies rate limiting to heavy use of the top models. The video's presenter reports he could not find published thresholds for when throttling starts or how slow responses can get, so treat "unlimited" as generous rather than unconditional.
How Max Mode billing works
Max Mode is Cursor's extended-context agent mode, billed separately from the flat Pro plan. Cursor charges you the underlying model provider's token cost plus a 20% margin, and expresses that charge in request equivalents: 90 requests equals one million input tokens for Claude 4, per the table shown in the video.
Everything sent to the model counts as billable input: your prompt, attached files, folders, and all retrieved context. Unlike normal mode, token volume directly drives cost in Max Mode, which is why the presenter warns against casual chat conversations while the mode is enabled.
The presenter's own session log shows the difference in practice: a normal-mode call with roughly 23,100 input tokens and 421 output tokens was marked "included in Pro" and cost him nothing, while a Max Mode call with Claude 4 produced a charge of about six cents for a small exchange.
What Max Mode actually gives you
Max Mode trades money for autonomy and context. Cursor advertises a much larger context window, file reads of up to 750 lines, and a higher tool-call ceiling, which lets a long coding task run without interruption.
In normal mode, the agent pauses after about 25 tool calls and asks you to resume. In Max Mode, the ceiling rises to roughly 200 calls per run, per the video. The presenter says that with a well-engineered prompt he can let the agent work through an entire task, including tests, before control returns to him. His caveat is direct: if your prompt is weak, you pay the Max Mode premium to watch the model wander.
Model costs per million tokens in Max Mode
The cost of Max Mode depends almost entirely on the model you pick. The figures below come from Cursor's pricing table as shown in the June 2025 video, expressed in request units where each request is valued at four cents.
Vendor pricing also changes frequently; check each provider's own page, such as Anthropic for Claude or Google's Gemini site for Gemini, before budgeting a large codebase task.
| Model | Input (requests per 1M tokens) | Cached input | Output (requests per 1M tokens) | Approx. cost per 1M in/out |
|---|---|---|---|---|
| Claude 4 | 90 | 9 | 450 | $3.60 / $18.00 |
| Claude | 450 | not stated | 2,250 | $18.00 / $90.00 |
| Gemini (long-context tier) | 75 + 18 surcharge | not stated | 75 + 18 surcharge | cheaper than Claude per video |
Cached input is the cheap row: repeated prompts that Cursor matches against prior context are billed at about 10% of the full input rate. The cache is managed automatically, so you benefit without configuring anything.
A worked cost example from the video
A concrete example makes Cursor pricing in Max Mode tangible. Suppose one message consumes 135,000 input tokens and produces 82,000 output tokens on Claude 4, using the June 2025 table.
- Input: 135,000 / 1,000,000 × 90 = 12.15 requests, about $0.49 at four cents per request.
- Cached input: billed at 9 requests per million, so a fully cached equivalent would cost 0.9 requests, roughly $0.04.
- Output: 82,000 / 1,000,000 × 450 = 36.9 requests, about $1.48.
- Total: roughly $2 for a single large exchange, before any additional tool calls.
Two rules summarize the arithmetic. Sending one million input tokens costs 90 requests, or $3.60, and receiving one million output tokens costs 450 requests, or $18. Output is five times the input rate because generation is more expensive for the provider than reading context.
How to spend less while using Max Mode
The biggest lever is model selection. The video shows that OpenAI's GPT-family models, served via OpenAI, and Google's Gemini are priced well below Claude in Max Mode, so reserving Opus for genuinely hard reasoning keeps costs down.
The presenter's second recommendation is to use Gemini-class models with very large context windows, reportedly up to one million tokens, when a task requires reading a big codebase or framework internals. The large window plus a lower per-token rate makes them the economical choice for wide-context analysis.
Finally, write prompts that return control only when the task is genuinely finished. With the higher tool-call ceiling in Max Mode, a precise prompt converts the pay-per-token model into long autonomous runs instead of many short, repeated, paid exchanges.
Caveats the video does not resolve
Three limits deserve emphasis. First, the rate-limiting rules for the unlimited Pro plan are not published, so heavy users cannot predict when responses slow down. Second, Cursor's own documentation can lag the live rollout, as the video demonstrates when the presenter's account shows both old and new plan states within hours. Third, token accounting in Max Mode depends on Cursor's caching behavior, which the company can optimize over time; exact per-request charges may not be fully reproducible by users.
The video was published on June 18, 2025. Pricing pages, model lineups, and request tables have very likely moved since then, so verify everything on the current Cursor pricing page before making a purchase decision.
CTA: Turn videos like this into written guides
This article exists because a 26-minute video about Cursor pricing contained numbers worth putting in writing: the request table, the 20% margin, the worked example. If you publish explanations like that on YouTube, the same content can become a searchable, quotable article.
Skala Blog turns a YouTube URL into a structured blog post: paste the link, the video is transcribed, and an article draft is generated for your review. The same flow works whether your channel covers developer tools, courses like Crazystack Typescript or the Bootcamp do Dev Doido from Gustavo Dev Doido, or any topic where your best material lives in video. Visit crazystack.com.br to see one such ecosystem, and Skala Blog to repurpose your own.
Frequently asked questions
- Is Cursor's Pro plan really unlimited since June 2025? Normal-mode requests, Tab completions, and background agents are marketed as unlimited on the $20 Pro plan as of June 2025. Cursor applies rate limiting to heavy use of top models, and the thresholds are not published.
- What does Max Mode cost? Max Mode is billed at the chosen model provider's token cost plus a 20% Cursor margin, expressed in request units. For Claude 4 in June 2025 that meant 90 requests per million input tokens and 450 per million output tokens, with each request valued at four cents.
- Do I pay per token in normal mode? No. In normal mode each message was a flat request and, after the June 2025 change, requests fall inside the Pro subscription. Token volume only affects billing in Max Mode.
- Why is output more expensive than input? Providers charge more to generate tokens than to read them because generation requires sequential computation. In the June 2025 Claude 4 table, output cost five times the input rate.
- Is cached input cheaper in Max Mode? Yes. Cursor matches repeated prompt content against prior context and bills it at roughly 10% of the standard input rate, managed automatically.
Fork this article
Start a new branch from the same video, shaped your way. You keep the credit; the original keeps the attribution.
A fork in another language is filed as a translation of this article, so the two pages point at each other. You can unlink it later from the editor.
0/240
You are creating
- Format
- For
- Language
- Source
- Your angle
You will be asked to sign in before it is generated.
Buy credits