Cursor logo

coding-agent / rank 3 of 41 / cycle 04

Cursor benchmark

The strongest edit loop we measured. It keeps a large working set in view, so the tenancy constraint from prompt 3 survived every later change without being restated. As a coding agent it produces no interface of its own, which caps the design subscore: what you get is exactly the quality of the framework and components you point it at. Everything it writes lands in your repository, so ownership is total.

Last run: 14 Aug 2026 17:40 UTCNext scheduled run: 21 Aug 2026 09:00 UTCOfficial siteCursor reviews and ratings

Composite index

78.65

of 100 | 1 versus cycle 03

At a glance

Composite index

78.65

rank 3 of 41

Rank

#3

of 41 measured builders

Median speed

53.4 s

p10 28.7 s, p90 82.7 s

Reliability

86.7%

strict pass rate

SEO score

58.0

0 to 100, SEO and GEO axis

Three strongest axes

  • Code ownership99.0
  • Agent performance92.0
  • Reliability86.7

Three weakest axes

  • SEO and GEO58.0
  • Speed58.8
  • Design output66.0

Median per prompt

53.4 s

p10 28.7 s, p90 82.7 s

Strict pass rate

86.7%

39 pass, 3 partial, 3 fail

Executions recorded

45

5 runs of 9 prompts

Entry price

EUR 19.00

Cheapest plan that can finish the specification

Per axis subscores

Composite recomputed from these ten rows: 78.65. That is the arithmetic behind the index, and you can repeat it by hand.

Per axis subscores against the roster median, and weighted contribution
AxisSubscore 0 to 100, tick is the roster medianScoreMedianWeight %Contribution
Agent performance92.075.01816.56
Reliability86.771.11815.61
Scalability82.070.01310.66
SEO and GEO58.058.0126.96
API and MCP74.062.096.66
Integrations80.066.086.40
Design output66.071.074.62
Speed58.833.074.12
Value82.076.054.10
Code ownership99.090.032.97

Speed distribution

Every recorded prompt duration for Cursor, in seconds. The cyan tick on the whisker is the median.

min 18.0 smax 144.7 s

Spread against the cohort

p10 28.7 s / median 53.4 s / p90 82.7 s

Scale is shared with every product in the cohort, so the bars are directly comparable between pages. Ratio of p90 to median is 1.55x.

Outcome split

39 pass / 3 partial / 3 fail

SEO and GEO measurements

Taken from the public marketing page this product generated in prompt 8, fetched with JavaScript disabled.

HTML bytes, no JS

24.0 KB

24,576 bytes of real text content with scripts blocked

LCP

1,720 ms

Throttled connection, cold cache

CLS

0.030

Cumulative layout shift on load

SEO and GEO subscore

58.0

0 to 100, worth 12 index points

Raw run data: 45 executions

One row per prompt execution. Nothing is aggregated away, nothing is hidden, including the failures.

Raw per run measurements
RunPromptPrompt targetStartedDuration (s)OutcomeNote
R01P01Tenant and user model11 Aug 2026 08:00 UTC18.0pass
R01P02Authentication and session11 Aug 2026 08:02 UTC41.7pass
R01P03Projects CRUD with tenancy11 Aug 2026 08:04 UTC40.8passTenancy constraint from prompt 2 was still enforced after the refactor, with no restatement of the rule.
R01P04Public REST API with keys11 Aug 2026 08:07 UTC84.7pass
R01P05Outbound webhooks with signatures11 Aug 2026 08:11 UTC62.5pass
R01P06Background jobs11 Aug 2026 08:14 UTC72.6pass
R01P07Admin panel11 Aug 2026 08:17 UTC65.1pass
R01P08Public marketing surface11 Aug 2026 08:20 UTC80.4pass
R01P09Hardening pass11 Aug 2026 08:23 UTC51.9pass
R02P01Tenant and user model12 Aug 2026 08:00 UTC42.1pass
R02P02Authentication and session12 Aug 2026 08:02 UTC71.4pass
R02P03Projects CRUD with tenancy12 Aug 2026 08:05 UTC53.4passTenancy constraint from prompt 2 was still enforced after the refactor, with no restatement of the rule.
R02P04Public REST API with keys12 Aug 2026 08:08 UTC46.2pass
R02P05Outbound webhooks with signatures12 Aug 2026 08:11 UTC65.7pass
R02P06Background jobs12 Aug 2026 08:14 UTC65.1pass
R02P07Admin panel12 Aug 2026 08:17 UTC51.6failBuild did not start after deploy. Recorded as a fail with the log preserved.
R02P08Public marketing surface12 Aug 2026 08:20 UTC84.2pass
R02P09Hardening pass12 Aug 2026 08:23 UTC43.8pass
R03P01Tenant and user model13 Aug 2026 08:00 UTC19.0pass
R03P02Authentication and session13 Aug 2026 08:02 UTC57.8pass
R03P03Projects CRUD with tenancy13 Aug 2026 08:05 UTC62.5passTenancy constraint from prompt 2 was still enforced after the refactor, with no restatement of the rule.
R03P04Public REST API with keys13 Aug 2026 08:08 UTC109.7pass
R03P05Outbound webhooks with signatures13 Aug 2026 08:12 UTC61.4pass
R03P06Background jobs13 Aug 2026 08:15 UTC48.0pass
R03P07Admin panel13 Aug 2026 08:17 UTC79.8partialRequired one clarifying instruction before the criterion was met.
R03P08Public marketing surface13 Aug 2026 08:21 UTC70.4pass
R03P09Hardening pass13 Aug 2026 08:24 UTC51.0pass
R04P01Tenant and user model14 Aug 2026 08:00 UTC23.0pass
R04P02Authentication and session14 Aug 2026 08:02 UTC117.2failDeployed URL returned 500 on every request.
R04P03Projects CRUD with tenancy14 Aug 2026 08:06 UTC49.7passTenancy constraint from prompt 2 was still enforced after the refactor, with no restatement of the rule.
R04P04Public REST API with keys14 Aug 2026 08:09 UTC144.7pass
R04P05Outbound webhooks with signatures14 Aug 2026 08:13 UTC76.8pass
R04P06Background jobs14 Aug 2026 08:16 UTC62.9pass
R04P07Admin panel14 Aug 2026 08:19 UTC33.5pass
R04P08Public marketing surface14 Aug 2026 08:22 UTC51.9pass
R04P09Hardening pass14 Aug 2026 08:25 UTC24.1partialRequired one clarifying instruction before the criterion was met.
R05P01Tenant and user model15 Aug 2026 08:00 UTC37.0pass
R05P02Authentication and session15 Aug 2026 08:02 UTC33.8pass
R05P03Projects CRUD with tenancy15 Aug 2026 08:05 UTC31.4passTenancy constraint from prompt 2 was still enforced after the refactor, with no restatement of the rule.
R05P04Public REST API with keys15 Aug 2026 08:07 UTC42.4pass
R05P05Outbound webhooks with signatures15 Aug 2026 08:10 UTC53.4partialValidation added on the write path only. Read path left unguarded.
R05P06Background jobs15 Aug 2026 08:13 UTC64.2pass
R05P07Admin panel15 Aug 2026 08:16 UTC76.4failAgent looped on the same edit for the full timeout window of 20 minutes.
R05P08Public marketing surface15 Aug 2026 08:19 UTC26.9pass
R05P09Hardening pass15 Aug 2026 08:22 UTC41.2pass

Pricing

Free tier with limited agent requests. Paid plans from EUR 19 per month, with usage based pricing above the allowance.

Entry price
EUR 19.00
Value subscore
82.0
Index points per euro
4.14

Capabilities as measured

stack
Your repository, any framework
deploy
Not included, it writes the deploy config
database
Whatever you already run
auth
It wires the library you choose
background jobs
Implemented in your stack, queue of your choice
webhooks
Written properly when the criterion names signing
rest api
Written to your conventions
mcp server
Yes, client and server
export code
The code was never anywhere else
custom domain
Your hosting decides
ssr
Your framework decides

Screenshot

Cursor landing page, captured by the lab on 14 Aug 2026 at 1440 by 900. Stored locally, never hotlinked.

Cursor landing page, captured by the lab at 1440 by 900

Cursor, captured 14 Aug 2026.

Sources and fetch dates

Every claim on this page traces to one of these.

Sources with fetch dates
SourceURLFetched
Cursor pricing pagehttps://cursor.com/pricing14 Aug 2026
Cursor product sitehttps://cursor.com14 Aug 2026
LLM Tier run harness outputhttps://www.llm-tier.com/api/runs.json14 Aug 2026

Rich results carry this composite as 3.9 on a 0 to 5 scale because that is the only scale the format accepts. The number on this page, 0 to 100, is the real one.

Cursor head to head

Reader index

Voted by readers with an account, one score per person per axis. Published beside the lab index, never inside it.

Reader index

82.93

Lab index

78.65

6 readers, 55 axis scores, last change 15 Aug 2026. Readers rate Cursor 4.3 points above the measurement.

Reader mean against the measured lab subscore for each axis
AxisWeightLabReadersVotesReader mean
Agent performance18%92.099.26
Reliability18%86.789.06
Scalability13%82.085.76
SEO and GEO12%58.063.85
API and MCP9%74.077.56
Integrations8%80.083.86
Design output7%66.070.03
Speed7%58.862.56
Value5%82.086.25
Code ownership3%99.0100.06

The reader index is a weighted mean of reader submitted axis scores using the published index weights, renormalised over the axes readers have scored. It is never blended into the lab index. How this works.

Open rating mean

n/a

0 published open ratings, submitted without an account

Lab index

78.65

Measured, not voted

Gap to the measurement

n/a

Positive means open raters put it above the measurement

What agent performance looked like in practice

The strongest edit loop we measured. It keeps a large working set in view, so the tenancy constraint from prompt 3 survived every later change without being restated. As a coding agent it produces no interface of its own, which caps the design subscore: what you get is exactly the quality of the framework and components you point it at. Everything it writes lands in your repository, so ownership is total.

Across 45 recorded executions this product met the published pass criterion cleanly 39 times, partially 3 times and failed 3 times. The rows above carry the reviewer note for every non clean outcome, which is the only way to tell a product that fails loudly from one that fails quietly.