Cursor API pricing now revolves around the new flagship Grok 4.6: $2 input / $6 output per 1M tokens at list, with a 50% launch-week discount from 2026-08-12; Composer 2.5 is the cheapest at $0.5/$2.5 and Auto routing sits at $1.25 input + cache write / $6 output (source). But first, get one thing straight: this is not a standalone OpenAI-style REST API—it's per-token billing that runs through your Cursor subscription pool via the SDK and IDE. This post walks through the rate table, caching, API key setup, integration, and whether it's worth it.
Core Pricing Comparison: Grok 4.6, Composer 2.5, and Auto
The public rates split into three tiers: the newest flagship Grok 4.6 at $2/$6 (50% off launch week), Composer 2.5 at the low end of $0.5/$2.5, and Auto routing in the middle at $1.25 input + cache write / $6 output; the previous Grok 4.5 matches Grok 4.6, and Fast tiers add a multiplier on top of the same models.
| Model | Input | Output | Cache Read | Notes |
|---|---|---|---|---|
| Grok 4.6 | $2 / M | $6 / M | $0.5 / M | Newest flagship, 50% off launch week |
| Grok 4.6 Fast | $4 / M | $12 / M | $1 / M | Flagship low-latency tier |
| Grok 4.5 | $2 / M | $6 / M | $0.5 / M | Previous flagship |
| Grok 4.5 Fast | $4 / M | $12 / M | $1 / M | Previous low-latency tier |
| Composer 2.5 | $0.5 / M | $2.5 / M | $0.2 / M | Budget mainline, recommended SDK id |
| Composer 2.5 Fast | $3 / M | $15 / M | $0.5 / M | Budget low-latency tier |
| Auto | $1.25 (Input+Cache Write) / M | $6 / M | $0.25 / M | Cursor auto routing |
Rates come from the official models-and-pricing docs (source), per 1M tokens; Grok 4.6 is 50% off during its launch week effective 2026-08-12.
Grok 4.6: Newest First-Party Flagship
Grok 4.6 is Cursor's newest first-party flagship at $2/$6 with $0.5/M cache reads, and Fast at $4/$12 for lower latency. With the launch-week discount from 2026-08-12, it's the model worth trying first in the Cursor pool right now; Grok 4.5 matches its price as the previous, more mature tier.
Composer 2.5: The Budget Mainline
Composer 2.5 is the low-cost agentic coding model in Cursor's pool at $0.5/$2.5 with $0.2/M cache reads, matching the recommended SDK model id composer-2.5. Daily agents, Cloud Agents, and Cursor SDK automation run on it by default—the cheapest mainline tier on the table.
Auto: Smart Routing
Auto lets Cursor pick models, priced at $1.25 input + cache write, $6 output, and $0.25 cache read. Choosing Auto, or letting Composer 2.5 serve as the Auto route target, draws the Auto + Composer pool—which official docs say is typically more allowance-friendly than manually picking third-party flagships.
Hidden Legacy Versions
Composer 1/1.5/2 stay hidden by default, and 1.5's $3.5/$17.5 is actually pricier than the 2.x line—watch for that when migrating old integrations.
Caching and Usage Pools: The Savings Live in the Pool, Not the Unit Price
Grok 4.6 cache reads run $0.5/M, Composer 2.5 just $0.2/M, and Auto cache reads $0.25/M, so repeated context should hit the cache first; but the real cost gap comes from usage pools—the Cursor models pool (Grok 4.6, Grok 4.5, Composer 2.5) draws the Auto + Composer pool, which is usually more generous than the API pool.
- Cache reads: Composer 2.5 $0.2/M, Grok 4.6 / Grok 4.5 $0.5/M, Auto $0.25/M (source)
- Usage pools: manual third-party picks draw the API pool; Auto / Composer 2.5 draw the Auto + Composer pool
- Teams Cursor Token Rate: third-party models and Auto Balance/Intelligence routes add $0.25/M; Auto Cost and all first-party Cursor models (Grok 4.6, Grok 4.5, Composer 2.5) are exempt
- Max Mode: legacy request-based plans only, at model API price plus 20%
- On-demand: after the pool empties, billing continues at the same rates—official notes say requests are never downgraded in quality or speed
API Key Setup, Integration, and Boundaries
Generate CURSOR_API_KEY in the Cursor Dashboard Integrations console and connect through the Cursor IDE, CLI, Cloud Agents, or Cursor SDK; it is not a standalone OpenAI-compatible REST endpoint, so don't treat it as a generic inference API.
- Entry: cursor.com/docs/models-and-pricing is the official pricing and reference, with purchases going through cursor.com/pricing
- API key: Dashboard → Integrations → generate CURSOR_API_KEY, not from a third-party platform
- Integration: manual model picks in the Cursor IDE, CLI, Cloud Agents, or the SDK (TypeScript @cursor/sdk, Python cursor-sdk) with model id composer-2.5
- Billing tie-in: no separate balance—usage draws the subscription's API pool or Auto + Composer pool, falling back to on-demand
- Rate limits / quotas: no fixed rate-limit number is published; check the console and official docs
Is It Worth It: Do the Math Together With Your Subscription Pool
Whether Cursor API is worth it depends on treating it as an extension of your subscription rather than a standalone API—the newest flagship Grok 4.6 is the main pick on a launch-week discount, with Composer 2.5's $0.5/$2.5 as the budget tier for cost-sensitive workloads; the real decision is the pool math, not the unit price alone.
| Workload | Recommended Billing | Rate Reference | Why |
|---|---|---|---|
| Try the newest flagship | Grok 4.6 | $2 / $6 / M (50% off launch week) | Latest capability at a launch window price |
| Daily agents, Cloud Agents | Grok 4.6 family / Auto | Grok 4.6 $2/$6 M; Auto Input+Cache Write $1.25 + Output $6 / M | Prefer the newest flagship inside the pool |
| SDK automation | Grok 4.6 family | $2 / $6 / M (50% off launch week) | Newest flagship, launch-window trial |
| Cost-sensitive / light tasks | Composer 2.5 | $0.5 / $2.5 / M | Official lowest-price budget tier |
| Mature and stable first | Grok 4.5 | $2 / $6 / M | Same rates as Grok 4.6, previous gen |
| Latency-first | Fast tiers per model | Official table multipliers | Trade cost for response speed |
Versus the subscription: Pro at $20/month already includes a $20 API pool, with Pro+ at $60 and Ultra at $200 aimed at daily agents and power users (see the Cursor plan). Running Grok 4.6 / Auto inside the subscription is cheaper than raw pay-as-you-go; only heavy SDK automation that exhausts the pool falls back to on-demand.
For cross-vendor comparisons, note the form factor first: OpenAI API (GPT-5.6 Luna at $0.20/$1.20) and Claude API are standalone open endpoints, while Cursor is a subscription-linked system—different shapes, so don't compare unit prices head-to-head.
Bottom Line
- The official pricing docs at https://cursor.com/docs/models-and-pricing are the single source of truth—don't miss the Grok 4.6 launch discount, Fast tiers, and the Teams Token Rate.
- Confirm which usage pool your requests draw from—the API pool and the Auto + Composer pool can differ a lot in cost at the same rates.
- Cursor API is not OpenAI-compatible; if you plan to use it from your own backend, check the integration path before you start.
- Subscription and API are one ledger: after the pool empties, on-demand continues at the same rates with no cliff in speed.