ARAgentRanks
~ / comparisons / cursor-grok-4-5-cost

Cursor + Grok 4.5 cost, shown as burn rate.

The useful question is not just "which Cursor tier?" It is how quickly Grok 4.5 burns value when you use standard, Fast, Max Mode, cache-heavy context, or retry loops.

Pick a tierSee token burnCursor review

Plan ladder

Official monthly price and API usage included. First-party pool size is not disclosed.

Pro$20/mo
$20 API
Pro Plus$60/mo
$70 API
Ultra$200/mo
$400 API

undisclosed Cursor's First-party models pool includes Auto, Composer 2.5, and Grok 4.5, but Cursor describes the included amount only as generous usage. Treat exact pool depth as unknown.

What the API budget buys

API-equivalent math using Grok 4.5 standard rates. This is not a claim about Cursor's undisclosed First-party pool.

Assumptions: total token est. uses a cache-heavy standard Grok 4.5 mix: 2% uncached input, 88% cache read, 10% output = about $1.08 per 1M blended tokens. Normal agent round = 100k fresh input + 400k cache + 15k output = $0.49. Small webpage pass = 2 normal rounds = $0.98. Image brief iteration = 20k input + 5k output = $0.07, excluding image-generation fees.

Pro

$20 API
Total token est.
18.5M blended tokens
Agent rounds
40 normal rounds
Web pages
20 small page passes
Image briefs
285 text-only iterations

Pro Plus

$70 API
Total token est.
64.8M blended tokens
Agent rounds
142 normal rounds
Web pages
71 small page passes
Image briefs
1,000 text-only iterations

Ultra

$400 API
Total token est.
370M blended tokens
Agent rounds
816 normal rounds
Web pages
408 small page passes
Image briefs
5,714 text-only iterations

Grok 4.5 burn rate

Prices per 1M tokens. Fast mode is the expensive path, especially on output.

Grok 4.5 standard

Input
$2
Cache
$0.5
Output
$6

Grok 4.5 Fast

Input
$4
Cache
$1
Output
$18

Same task, different burn

These are transparent token math examples, not claimed Cursor pool limits.

Normal agent round

Tokens
100k fresh input + 400k cache + 15k output
Standard
$0.49 per round, about 40 rounds per $20
Fast
$1.07 per round, about 18 rounds per $20

Large codebase round

Tokens
200k fresh input + 1M cache + 40k output
Standard
$1.14 per round
Fast
$2.52 per round

Long cached session

Tokens
50k fresh input + 2.5M cache + 50k output
Standard
$1.65 per round
Fast
$3.60 per round

Max-context style round

Tokens
500k fresh input + 50k output, no cache assumed
Standard
$1.30 per round
Fast
$2.90 per round

Which tier fits the work?

Use the cheapest tier that covers your real failure cost, not the tier that sounds safest.

Occasional Grok testPro is enough. You are validating behavior, not running daily long tasks.
Light daily vibe codingPro can work if you avoid Fast and Max Mode by default.
Daily agent projectPro Plus is the practical middle tier for repeated coding sessions.
Multi-agent automationUltra is the realistic tier when parallel agents and long tasks are normal.
Cost controlUse standard Grok 4.5 first. Fast output is 3x the standard output price.

AgentRanks take

Pool opacity is the buying risk.

Grok 4.5 standard looks reasonable per token. The hard part is Cursor's First-party models pool: users can see the tier, but not a clean published token allowance. For pricing pages and rankings, AgentRanks should show the known burn rate and mark the hidden pool limit as unknown.

Sources

Last checked 2026-07-14.