Cursor Start is a new subscription tier for developers in India that costs ₹649 per month (about $7.80 USD), tax-inclusive, payable via UPI or card. Launched on July 28, 2026, it sits between Cursor's Free plan and the $20/month Pro plan, and it includes generous access to Grok 4.5 and Composer 2.5 — two models that would cost roughly ten times more per token if you used them through frontier API providers like Anthropic or OpenAI.
The question is not whether ₹649 is cheap. At roughly one-third of Pro's dollar price, it obviously is. The question is whether the plan gives you enough to actually build with every day, or whether the things it leaves out — third-party models, Bugbot, Fast mode, on-demand overages — make it a dead end for serious work.
After verifying the plan details against Cursor's official announcement, the pricing page, and independent reporting, here is what Start actually includes, what it doesn't, and who should buy it.
What Cursor Start Includes
Cursor Start gives developers in India expanded access to Cursor's own models at a price built for the local market. According to Cursor's official announcement, the plan includes:
- Grok 4.5 — Cursor's most powerful model, trained jointly with SpaceXAI and released on July 8, 2026. It scores 83.3% on Terminal-Bench 2.1 and 64.7% on SWE-Bench Pro, putting it in the same class as Opus 4.8 and GPT-5.5 for coding work, at $2 per million input tokens and $6 per million output tokens on the API.
- Composer 2.5 — Cursor's most price-efficient coding model, released May 18, 2026. It is built on Moonshot AI's Kimi K2.5 checkpoint with heavy Cursor-specific RL, and scores near Opus 4.7 on SWE-Bench Multilingual while costing $0.50 per million input tokens and $2.50 per million output tokens on the standard tier — roughly 10x cheaper than frontier models.
- More agent requests than the Free plan, across desktop, web, iOS, and the CLI.
- Always-on cloud agents — long-running agents that build, test, and open pull requests while you keep working locally.
- Cursor for iOS — launch agents or control existing ones from your phone, then pick up the work on your desktop.
- Plugins, MCP servers, hooks, and skills — the extension ecosystem that makes Cursor adaptable to your workflow.
- Local pricing — ₹649 per month, tax-inclusive, billed monthly in INR with UPI or card.
That is a substantial package for ₹649. To put it in perspective: if you ran the equivalent volume of Grok 4.5 and Composer 2.5 tokens through the xAI and Cursor APIs directly, you would burn through ₹649 in a few hours of heavy agent use. Start subsidizes that cost into a flat monthly fee.
What Start Leaves Out (and Why It Matters)
Cursor is explicit that Start is "not a discounted Pro." The plan covers everyday building on Cursor's own models. When you need more, you upgrade to Pro. The specific things Start does not include:
No third-party models. Start gives you access to Cursor's own models — Grok 4.5 and Composer 2.5. It does not include the "Other Models" pool that Pro offers, which means no Claude Opus, no GPT-5.5, no Gemini, no third-party model routing. If your workflow depends on a specific frontier model from another lab, Start will not cover it.
No Bugbot. Cursor's agentic code review tool stays locked to Pro and above. If you want automated code review on your pull requests, you need Pro.
No Auto mode or Automations. The auto model selection feature (also known as Cursor Router, which picks the best model for each task automatically) and Automations (scheduled or triggered agent workflows) are Pro features.
No Cursor SDK. Programmatic agent building — constructing agents in code rather than through the IDE — requires Pro.
No on-demand overages. If you hit your Start usage ceiling, you cannot pay extra to keep going. You either wait for the next billing cycle or upgrade to Pro. Pro includes on-demand usage that lets you continue past included limits with usage-based billing.
No Fast mode or effort controls. On Start, Grok 4.5 runs at medium effort, and Composer runs in non-fast mode. Pro lets you dial effort up or down and toggle Fast mode for lower latency. This matters more than it sounds: Fast mode is the default in Cursor's model picker for a reason — it trades a small quality dip for a significant speed gain on long agent sessions.
Free vs. Start vs. Pro: The Honest Comparison
Here is how the three individual plans stack up, based on Cursor's pricing page and the official Start announcement:
| Feature | Free | Start (India) | Pro |
|---|---|---|---|
| Price | ₹0 | ₹649/mo (tax inclusive) | $20/mo (before tax) |
| Payment | n/a | UPI or card | Card / global billing |
| Grok 4.5 | No | Generous access | Generous access |
| Composer 2.5 | Limited | Generous access | Generous access |
| Third-party models (Claude, GPT, Gemini) | No | No | Yes |
| Cloud agents | Limited | Included | Included |
| Cursor for iOS | Limited | Included | Included |
| MCP / plugins / hooks / skills | Basic | Yes | Yes |
| Bugbot | No | No | Yes |
| Auto mode / Automations | No | No | Yes |
| Cursor SDK | No | No | Yes |
| Fast mode / effort controls | n/a | Locked (medium effort) | Configurable |
| On-demand overages | No | No | Yes |
The pattern is clear: Start gives you the Cursor-native building experience at a local price. Pro gives you the full model menu and the power tools. The gap between them is not about usage volume — it is about model choice and control.
Who Should Buy Cursor Start
Buy Start if you:
- Are a developer in India currently on the Free plan who burns through local agent limits mid-week.
- Want cloud agents and iOS steering without the $20/month USD friction.
- Build primarily with Cursor's own models (Grok 4.5 and Composer 2.5) and do not need Claude, GPT, or Gemini inside the IDE.
- Need UPI or INR billing — students, freelancers, and early-stage startups who cannot easily pay in USD.
Stay on Free if you:
- Only need occasional Composer access or light tab completions.
- Are evaluating Cursor for the first time and want to test it before committing.
Jump to Pro (or stay on Pro) if you:
- Require Claude, GPT, or multi-lab model routing inside the IDE.
- Use Bugbot for agentic code review on pull requests.
- Need Auto mode, Automations, or the Cursor SDK in production.
- Want Fast mode and effort tuning for hard, long-running agent loops.
- Need on-demand usage past included limits without upgrading plans.
The Grok 4.5 and Composer 2.5 Advantage
The reason Start is a compelling deal at ₹649 — and not just a cheap teaser — is the quality of the models it includes. Both Grok 4.5 and Composer 2.5 are 2026 releases that compete with frontier models at a fraction of the cost.
Grok 4.5 was released by SpaceXAI on July 8, 2026, as xAI's first model trained from the ground up for coding and agentic tasks. It was co-trained with Cursor using real developer session data. On Terminal-Bench 2.1 it scores 83.3%, within 1 point of GPT-5.5 and just behind Claude Fable 5 at 84.3%. Its key advantage is token efficiency: on SWE-Bench Pro, Grok 4.5 uses approximately 15,954 output tokens per task, compared to Opus 4.8's 67,020 — a 4.2x gap that translates directly to lower cost per task on long agent runs. Musk called it "Opus-class, but faster," and the benchmarks largely support that framing: it is not the outright best model in any single category, but it may be the best value for high-volume agentic coding.
Composer 2.5 was released by Cursor on May 18, 2026, as the latest iteration of its proprietary coding model family. It is built on Moonshot AI's Kimi K2.5 open-weight checkpoint with heavy Cursor-specific reinforcement learning — 85% of its compute goes into additional training and RL beyond the base model. On SWE-Bench Multilingual it scores 79.8%, within 1 point of Opus 4.7 at 80.5%. On CursorBench v3.1 it scores 63.2%, above Opus 4.7 at default settings. The standard tier costs $0.50 per million input tokens and $2.50 per million output tokens — roughly 10x cheaper than Opus 4.7 and GPT-5.5 on a per-token basis. For long-running agent sessions where cost compounds over hours, that price gap is the difference between a viable workflow and one that burns your budget in an afternoon.
The practical takeaway: if you are building primarily with Cursor's own models, Start gives you frontier-adjacent coding intelligence for ₹649/month. If you need the absolute top of the benchmark leaderboard — Claude Fable 5 on SWE-Bench Pro at 80.4% — you need Pro or an API account with Anthropic.
The SpaceX Acquisition Context
Cursor Start does not exist in a vacuum. It launched weeks before the expected closing of SpaceX's ~$60 billion all-stock acquisition of Anysphere (Cursor's parent company), which was announced on June 16, 2026. SpaceX and xAI finalized their merger on May 6, 2026, giving the combined entity access to the Colossus supercluster — one of the largest AI training compute clusters in the world.
The acquisition matters for Start in two ways:
Model access. Grok 4.5 is an xAI model, and Composer 2.5 is being co-developed with SpaceXAI on Colossus 2. The acquisition tightens the integration between Cursor's IDE and the xAI model stack, which means Start users are getting first-class access to models that external competitors will not necessarily have.
India strategy. SpaceX already has a presence in India through Starlink. Cursor's India user base has tripled in the past year to more than 3 million developers, making it Cursor's third-largest market globally. India also has the highest agent requests per developer of any country. The Start plan is a commercial move to right-size pricing for that market, and the SpaceX infrastructure footprint in India could accelerate local deployment after the deal closes.
Cursor was running at approximately $4 billion ARR as of early June 2026. Start is not going to move that number dramatically in the short term — it is a ₹649 plan in a market where most users were on Free — but it is a land-grab play. Every developer who upgrades from Free to Start is a paying customer who is less likely to churn to a competitor.
For a deeper look at the acquisition and what it means for developers, see our earlier coverage of SpaceX's $60 billion Cursor acquisition and how Grok 4.5 fits into the enterprise AI price war.
How to Get Cursor Start
- Confirm you are signing in as an India user — your account country and payment method must both match India.
- If you are an existing Free user, go to your Cursor dashboard and upgrade to Start. If you are new, sign up at cursor.com/signup and select the Start plan during onboarding.
- Pay ₹649 with UPI, credit card, or debit card.
- Verify in settings that you are on the Cursor Models pool (Grok 4.5 and Composer 2.5). Third-party models will show $0 until you move to Pro.
If Start does not appear on your pricing page: disable any VPN, confirm your India billing profile, and refresh cursor.com/pricing. Cursor has stated that Start is India-only and uses verification and anti-VPN checks to prevent users outside India from accessing the plan.
The Security Tip From the Video
One detail worth flagging from the YouTube review of the Start plan: the recommendation to set a minimum-release-age of one week in your pnpm configuration. This prevents Cursor's agent (or any npm-dependent tool) from automatically installing newly published NPM packages, which is a common vector for supply-chain attacks. A one-week delay gives the community time to detect and report malicious packages before they reach your node_modules.
Add this to your .npmrc or pnpm config:
minimum-release-age=1w
This is not a Cursor-specific tip — it applies to any agentic coding tool that installs dependencies — but it is especially relevant when you are running cloud agents that build and test code autonomously, which Start includes.
The Bottom Line
Cursor Start at ₹649/month is a genuine mid-tier for India, not a stripped-down Pro. It removes the Free-to-Pro cliff for daily agent work on Grok 4.5 and Composer 2.5, and the inclusion of UPI billing removes a real friction point for Indian developers who could not easily pay in USD.
The trade-off is clear and honest: you get Cursor's own models and the full agent experience (cloud agents, iOS, CLI, plugins) at a local price, but you lose third-party model access, Bugbot, Auto mode, the SDK, Fast mode, and on-demand overages. If your workflow is "I live in Claude or Codex inside the IDE," Start will feel incomplete. If your workflow is "ship with Cursor's agents all day," ₹649 is the intended fit — and at roughly $7.80 USD, it is one of the most aggressive purchasing-power-parity prices in the AI tooling market today.
For builders wondering how Grok 4.5 got here, see our guide on Grok 4.5's universal rollout and what Grok 4.6 and 4.7 mean for the road ahead. For the wider question of whether open-weight or closed models win the next phase of AI coding, our builder's decision framework breaks down the trade-offs.
Discussion
0 comments