The median Cursor developer added 176 lines of code a week on 1 January 2025. By 16 May 2026 that figure was 712.46. This post collects verified Cursor AI statistics on output, model costs, code review, enterprise use and market position, taken from Cursor’s own Spring 2026 report, Stack Overflow’s April 2026 pulse survey, Ramp spend data and company filings.
Key Cursor AI Statistics
How Much Code Do Cursor Developers Write?
Mean lines added per developer per week rose from 3.6K on 1 January 2025 to 8.6K on 16 May 2026. Cursor describes this as coding speed doubling year over year.
The median moved further in proportional terms than the mean. Cursor notes the metric is imperfect and treats it as a directional baseline rather than a productivity measure.
The 90th percentile and the median both climbed. Cursor reports p90 developers pulling farther from median developers in absolute lines added per week: 2.5K against 176 on 1 January 2025, and 8.8K against 712.46 on 16 May 2026.
| Metric | 1 Jan 2025 | 16 May 2026 |
|---|---|---|
| Lines added per developer per week, mean | 3.6K | 8.6K |
| Lines added per developer per week, p50 | 176 | 712.46 |
| Lines added per developer per week, p90 | 2.5K | 8.8K |
| Lines added per PR, p75 | 125.86 | 345.02 |
| Merged PRs with 1,000+ lines changed | 8% | 13.8% |
Source: Cursor Developer Habits Report, Spring 2026, covering 1 January 2025 to 16 May 2026. Excludes users on Privacy Mode.
Agent sessions also ran deeper. Mean tool calls per session went from 113.63 on 1 March 2026 to 145.08 on 16 May 2026, a rise Cursor puts at roughly 30% over two months.
Cursor AI Statistics On Power Users
Usage concentrates at the top of the distribution. Cursor reports Gini scores of 0.77 for AI lines written, 0.75 for AI spend and 0.72 for tokens consumed, on a 0-to-1 scale where higher means more concentrated.
Lines are more concentrated than tokens. P99 developers produce 46x more AI lines per day than the median active user, but merge 15x more PRs than the median active PR author. Similar concentration shows up across the wider market in these AI agent adoption figures.
| Ratio | AI lines/dev/day (MA7) | Merged PRs/dev/wk |
|---|---|---|
| p99 vs p50 | 46x | 15x |
| p90 vs p50 | 10x | 4x |
Source: Cursor Developer Habits Report, Spring 2026, data to 16 May 2026.
Cursor AI Statistics By Model Cost
Cost per agent request spans nearly 9x across the seven model families Cursor benchmarked. Cost per accepted added line spans roughly 7x, so pricier models recover part of the difference by producing more code that survives.
| Model | Mean cost per agent request | Mean cost per accepted line |
|---|---|---|
| Opus 4.7 | $1.57 | 1.10¢ |
| Opus 4.6 | $0.86 | 1.19¢ |
| GPT-5.5 | $0.81 | 1.09¢ |
| GPT-5.4 | $0.46 | 0.54¢ |
| Sonnet 4.6 | $0.44 | 0.54¢ |
| GPT-5.3 Codex | $0.30 | 0.56¢ |
| Composer 2.5 | $0.18 | 0.18¢ |
Source: Cursor Developer Habits Report, Spring 2026.
The two columns rank differently. Opus 4.6 costs less per request than Opus 4.7 but more per accepted line.
On CursorBench 3.1, Cursor’s internal eval suite, Composer 2.5 scored 63.2% at $0.55 a task. Opus 4.7 at maximum reasoning scored 64.8% at $11.02. Pricing across the wider tool market is tracked in these enterprise AI budget numbers.
Cursor AI Market Share Statistics Against Copilot And Claude Code
Stack Overflow’s April 2026 pulse survey asked 1,100 developers and working professionals which coding agent tools they had used in the past year for work. Cursor tied with OpenAI Codex on 22%, behind GitHub Copilot and Claude Code.
| Tool | Have used in past year for work |
|---|---|
| GitHub Copilot | 41% |
| Claude Code | 26% |
| OpenAI Codex | 22% |
| Cursor | 22% |
| Gemini/Code Assist | 21% |
| Google Antigravity | 18% |
| JetBrains AI | 14% |
Source: Stack Overflow April 2026 Pulse Survey, published 27 May 2026, 1,100 respondents, opt-in. Multi-select question, so shares do not total 100%.
One caution on that survey: the prose in the same Stack Overflow post quotes different shares over a six-month window, putting GitHub Copilot at 61%, Claude Code at 51%, and Codex and Cursor at 20% each. The table above uses the figures published in the post’s coding agent tools chart. Stack Overflow has not reconciled the two sets.
Ramp measures something different again. Its adoption rate counts the share of businesses buying from a vendor out of all businesses that bought in that software category over the past 12 months. As of June 2026, 95% of organizations with a Code AI vendor bought Cursor, ranking it first in the category, up by under one percentage point year on year.
Ramp also reports that 19.8% of businesses bought from the Code AI category at all in May 2026, down 1.4 points on the month. Invoices and survey responses are not the same measurement, and the two datasets cannot be combined. Broader tool-level usage sits in these workplace AI tool statistics.
Cursor AI Enterprise Statistics
Cursor’s June 2026 press release and its enterprise page give different figures for the same metrics, two months apart. Both are company-issued and Cursor has not explained the gap.
| Metric | Press release, 9 Jun 2026 | Enterprise page, 13 Aug 2026 |
|---|---|---|
| Fortune 500 companies using Cursor | 67% | 64% |
| Businesses on the platform | 50,000+ | 50,000+ |
| Enterprise lines of code per day | 150M+ | 100M+ |
Source: Cursor via Business Wire, 9 June 2026; Cursor enterprise page, checked 13 August 2026.
Nokia’s Chief Technology and AI Officer, Pallavi Mahajan, put more than 20,000 Nokia engineers on Cursor in that release. Rippling’s CTO reports adoption growing from 150 to over 500 engineers, about 60% of the org.
Coinbase
Coinbase is the named customer with the most published before-and-after data. Over 2,400 of its developers use Cursor, and 75% of all its PRs are created by agents across cloud and local.
| Coinbase metric | Figure |
|---|---|
| Developers using Cursor | 2,400+ |
| PRs created by agents | 75% |
| Increase in PRs merged per engineer since start of 2026 | 55% |
| Manual coding saved per developer per week | 7 hours |
| Idea to first PR | 8 days to under 30 minutes |
Source: Cursor customer story, 23 June 2026.
The headline 90% reduction, from 20 days to 1.8 days for idea to production, applies to one team under Senior Director of Engineering Chintan Turakhia, not to Coinbase as a whole.
Cursor AI Code Review Statistics
Bugbot ran through 11 versions and 40 major experiments between July 2025 and January 2026. Its resolution rate, measured by AI at PR merge time, went from 52% to over 70% in that window.
Cursor’s May 2026 pricing update gives a newer figure: 80% of identified bugs resolved by merge time. A high-effort setting finds 35% more bugs while resolution holds at 80%.
| Bugbot metric | V1, Jul 2025 | V11, Jan 2026 |
|---|---|---|
| Resolution rate | 52% | Over 70% |
| Bugs flagged per run | 0.4 | 0.7 |
| Resolved bugs per PR | ~0.2 | ~0.5 |
Source: Cursor engineering blog, 15 January 2026.
Resolved bugs per PR more than doubled, in Cursor’s own framing. Bugbot reviewed more than two million PRs a month as of January 2026 for customers including Rippling, Discord, Samsara, Airtable and Sierra AI. An average run costs $1.00 to $1.50.
Context, Caching And Automation
Cache reads dominate token activity. On 9 May 2026 they made up 89.9% of total tokens, with output at 0.6%.
The input-to-output token ratio climbed from 4.52x on 1 January 2026 to 11.41x on 9 May 2026. Models read far more than they write.
Automation moved in step. The share of agent changes reaching commits without a separate manual diff acceptance step rose from 7% on 1 January 2026 to a peak of 38.5% on 7 May, settling at 36.3% on 16 May. Cursor calls this more than a 5x increase since the start of the year.
AI-generated code also lasted longer. The share of accepted AI lines still present after 60 minutes rose from 76.6% on 7 January 2026 to 80.6% on 16 May 2026. Comparable adoption curves appear in these generative AI adoption rates.
Who Owns Cursor In 2026?
SpaceX announced on 16 June 2026 that it had signed a merger agreement to acquire Anysphere, Cursor’s parent, in an all-stock deal valued at $60 billion. The transaction was disclosed in a Form 8-K.
SpaceX secured the option in April 2026 and exercised it in June. The deal is expected to close in the third quarter of 2026, subject to regulatory approvals, and had not closed as of 13 August 2026. Context on the buyer’s own AI business sits in these Grok usage figures.
FAQs
How many lines of code does the median Cursor developer write?
The median Cursor developer added 712.46 lines per week on 16 May 2026, up from 176 on 1 January 2025, per Cursor’s Spring 2026 Developer Habits Report. The mean rose from 3.6K to 8.6K.
What is the cheapest model in Cursor?
Composer 2.5 averaged $0.18 per agent request and 0.18¢ per accepted added line in Cursor’s Spring 2026 report, the lowest of the seven model families benchmarked on both measures.
Do the Cursor AI statistics on Reddit match the published data?
Neither Cursor nor Stack Overflow publishes Reddit-derived usage data, so no figure here comes from Reddit. Every number above traces to Cursor’s own reports, Stack Overflow’s April 2026 pulse survey, Ramp, or company filings.
How many companies use Cursor?
Cursor reports over 50,000 businesses on its platform. Fortune 500 penetration was stated as 67% in its 9 June 2026 press release and 64% on its enterprise page in August 2026.
Is Cursor bigger than GitHub Copilot?
Not on survey headcount. Stack Overflow’s April 2026 pulse survey put GitHub Copilot at 41% and Cursor at 22%. Ramp’s invoice data ranks Cursor first in its Code AI category at 95% adoption.
Sources
https://cursor.com/insights
https://stackoverflow.blog/2026/05/27/agents-on-a-leash-agentic-ai-remains-mostly-monitored-at-work/
https://cursor.com/blog/building-bugbot
https://cursor.com/blog/coinbase
