GitHub Copilot statistics (2026)
GitHub Copilot statistics (2026)
You are putting together a budget case or an engineering brief, and a single number can change the recommendation. The trouble with GitHub Copilot statistics is that paid subscribers, all-time users, survey awareness, and workplace adoption are routinely presented as though they measure the same audience.
The short answer: Copilot has more than 4.7 million paid subscribers as of January 2026, but the latest independent developer survey puts its workplace adoption at 21% in May–July 2026, down from 29% a year earlier. The useful way to read the data is not as one growth curve, but as separate measures of reach, use, output, and trust.
Table of contents
Get the Mobile Testing Playbook Used by 800+ QA Teams
Discover 50+ battle-tested strategies to catch critical bugs before production and ship 5-star apps faster.
Key GitHub Copilot statistics in 2026
Keep the labels with these figures. Each answers a different question.
More than 4.7 million paid subscribers: Microsoft disclosed this total on its January 28, 2026 FY2026 Q2 earnings call, a 75% year-over-year increase. This is a paid-subscriber count, not an active-user total. Microsoft’s FY2026 Q2 earnings call is the primary disclosure.
20 million all-time users: Microsoft reported this cumulative count on July 30, 2025. GitHub told TechCrunch that the figure means all-time users, so it is not a DAU or MAU measure. See Microsoft’s FY2025 Q4 earnings call and TechCrunch’s clarification.
21% workplace adoption: In JetBrains’ Developer Ecosystem Survey 2026, fielded May–July 2026 among more than 15,000 professional developers worldwide, 21% said they used GitHub Copilot at work. That was down from 29% a year earlier. JetBrains’ 2026 survey is survey evidence, not a subscriber count.
26.08% more completed tasks: Across three randomized field experiments involving 4,867 developers at Microsoft, Accenture, and an anonymous Fortune 100 company, access to the AI tool increased completed tasks by 26.08% (standard error: 10.3%). Microsoft Research’s study summary reports the combined result.
180% more commits versus 30% more releases: A May 2026 NBER matched event study of more than 100,000 GitHub developers found this gap for autonomous coding agents. It is evidence about generations of AI coding tools, not a Copilot-only metric. NBER Working Paper 35275 reports the production hierarchy.
46% distrust AI-output accuracy: Stack Overflow’s 2025 Developer Survey received more than 49,000 responses from 177 countries. Its July 2025 release says 46% of developers did not trust AI-output accuracy, up from 31% in 2024. Stack Overflow’s survey release is the source for that year-over-year comparison.
How many people actually use Copilot
Paid subscribers are the current disclosed scale metric
Microsoft said GitHub Copilot had more than 4.7 million paid subscribers in January 2026. The company also said Copilot Pro+ subscriptions for individual developers grew 77% quarter over quarter. That tells you how many paying subscriptions Microsoft disclosed; it does not establish how many people used Copilot on a particular day or month.
The 20 million figure is broader and older. Microsoft reported 20 million users in July 2025, while GitHub clarified that the number was all-time users. You should not subtract 4.7 million from 20 million, turn either figure into a conversion rate, or call the 20 million figure an active-user base. They use different populations and different dates.
DAU and MAU are not public company-wide totals
GitHub defines IDE daily active users as unique users who interacted with Copilot each day, and IDE weekly active users as unique users active over a rolling seven-day period. Those are metrics available to enterprise administrators in usage reporting, not a public company-wide DAU or MAU disclosure. GitHub’s metric definitions explain the terms.
If your decision depends on current engagement, paid subscriptions and all-time users are incomplete proxies. The public record does not provide a current global DAU or MAU number to fill that gap.
Copilot workplace adoption in 2026
The adoption picture changed quickly in 2026. JetBrains’ January 2026 AI Pulse survey of more than 10,000 professional developers across eight languages found 76% awareness of GitHub Copilot and 29% workplace use. Among respondents at companies with more than 5,000 employees, workplace adoption was 40%. JetBrains’ April 2026 report documents that earlier snapshot.
By the May–July 2026 Developer Ecosystem Survey, Copilot workplace adoption had fallen to 21%, while Claude Code reached 39%, up from 18% in January. Copilot awareness remained 79%, including 86–90% across Europe, the UK, and the US. The same survey found 90% of professional developers used AI coding agents at work at least weekly and 68% used them daily.
That does not contradict subscriber growth. A paid-subscriber disclosure counts a different population from a survey’s self-reported workplace-use measure. It does mean you should avoid treating the 4.7 million subscriber total as proof that Copilot led every measure of developer use in 2026.
JetBrains sells its own AI products, including JetBrains AI and Junie. Its published survey describes a globally representative sample and methodology, but the publisher is not neutral; use the results as survey evidence rather than as a market census.
What productivity studies actually found
A 2022 controlled experiment measured faster task completion
A controlled GitHub Copilot experiment recruited 95 developers: 45 received Copilot and 50 did not. Among the 35 developers in each group who completed an HTTP-server JavaScript task, the Copilot group finished 55.8% faster: 71.17 minutes versus 160.89 minutes. The time-to-completion difference had a p-value of 0.0017 and a 95% confidence interval of 21% to 89%.
The completion-rate result requires a different label. The treated group’s completion rate was seven percentage points higher, but that difference was not statistically significant; its 95% confidence interval was -0.11 to 0.25. Peng, Kalliamvakou, Cihon, and Demirer’s paper separates the two findings. The experiment was conducted in 2022 and published in 2023, so it is historical experimental evidence—not a 2026 production benchmark.
Field experiments measured completed tasks
The larger field-experiment result is more representative of workplace conditions, but its outcome is still specific: completed tasks. The combined 26.08% increase across 4,867 developers does not measure bug density, test-pass rate, maintenance cost, or release reliability.
This distinction matters when you translate an AI-assistant result into an engineering goal. A task-completion gain can improve throughput while leaving review, integration, testing, and deployment as separate constraints.
More code is not more shipped software
NBER Working Paper 35275 makes that downstream constraint visible. Its May 2026 study combined AI-usage telemetry with a matched event study of more than 100,000 GitHub developers. It found cumulative commit increases of 40% for autocomplete, 140% for interactive coding agents, and 180% for autonomous coding agents.
For autonomous agents, the 180% increase in commits attenuated to 50% more projects and 30% more actual releases. The authors identify human bottlenecks—including review, coordination, and deployment—as constraints on the translation from code activity to shipped software.
The study does not say that GitHub Copilot creates a particular defect rate. It does provide a useful decision rule: do not use commit volume as your outcome metric when your objective is reliable releases.
Your QA strategy needs to make that distinction operational. The State of QA Automation 2026 report examines the testing-side pressures created by faster release cycles, while the vibe coding statistics report covers broader evidence on AI-assisted development and verification risks.
Adoption is up while trust is down
Stack Overflow’s 2025 survey found that 84% of respondents used or planned to use AI tools in development, up from 76% in 2024. The survey reports 47.1% daily use overall and 51% among professional developers. This is AI-tool adoption across the survey population, not GitHub Copilot adoption.
Among the 2,689 respondents who used or developed AI agents, 67.9% named GitHub Copilot as an out-of-the-box agent, copilot, or assistant; ChatGPT was 81.7%. The survey’s AI results describe that respondent subgroup rather than global market share.
Usage and trust moved in different directions. The 46% distrust figure does not tell you that AI outputs are wrong 46% of the time. It tells you that nearly half of surveyed developers expressed a lack of trust in their accuracy, which is a signal to preserve review and test controls as adoption expands.
The QA statistic nobody has published
No public statistic measures whether Copilot-assisted code passes QA. There is no published Copilot-specific figure for bug density, defect rate, regression count, or test-pass rate—and no mobile-app equivalent that separates generated code from other code.
The closest production-scale enterprise evidence comes from ZoomInfo’s rollout to more than 400 geographically dispersed developers, beginning in September 2023 and published in February 2025. ZoomInfo reported a 33% average suggestion-acceptance rate, 20% of lines of code accepted, and 72% developer satisfaction. Developers also identified a lack of domain-specific logic and inconsistency in code quality as limitations that required additional scrutiny. ZoomInfo’s engineering case study is a first-party account from the company using the product.
Those acceptance figures measure use, not quality. Read alongside the NBER shipping gap, they show why generated code still needs scrutiny before release—but neither source supplies the QA-side number a buyer would need to quantify the risk.
No first-party Quash data covers GitHub Copilot adoption, productivity, or generated-code quality. The missing measurement is especially relevant in mobile work, where a release can pass a narrow development workflow yet fail across device states and operating-system conditions. If you are building a mobile verification plan, Appium usage and adoption data provides context on one widely used automation ecosystem; it does not fill the missing Copilot quality metric.
Methodology and sources
This report prioritizes primary disclosures, original surveys, first-party deployment evidence, and research papers. It excludes pricing, revenue, market-size estimates, and forecasts because they do not answer the core statistics question with an equivalent verified primary measure here.
Use these terms precisely:
Paid subscribers are paying subscriptions disclosed by Microsoft as of January 2026.
All-time users are people who have used Copilot at some point, as reported in July 2025; they are not monthly or daily active users.
Awareness is the share of a survey sample that has heard of a tool.
Workplace adoption is the share of a survey sample reporting use at work.
Completed tasks, commits, projects, and releases are separate outcomes. None can substitute for a code-quality metric.
The survey figures have their own populations and fielding periods. Microsoft’s subscriber disclosure is a company-reported commercial metric; JetBrains and Stack Overflow provide survey evidence; academic and institutional studies measure defined experimental outcomes. Keeping those evidence types separate is the difference between a citable Copilot statistic and a misleading one.
Conclusion
For a defensible GitHub Copilot statistic, start with the question you need to answer. Cite 4.7 million for January 2026 paid subscribers, 20 million for July 2025 all-time users, and 21% for workplace adoption in JetBrains’ May–July 2026 survey. If your decision is about delivery outcomes, do not stop at a productivity or commit figure: define the QA and release metric you need before you decide what success looks like.
Continue reading
More articles from the Quash editorial desk.
