GitHub Octoverse 2026 counts 17.9 billion pushes as agents start more than 40% of pull requests
GitHub has published Octoverse 2026, its annual report on activity across the platform, and the headline figures describe a year in which the volume of software work on GitHub rose far faster than the number of people doing it. Developers pushed code 17.9 billion times between October 1, 2025 and September 30, 2026, more than in 2024 and 2025 combined. Commits reached 34.3 billion, more than three times the level of two years earlier, merged pull requests reached 1.3 billion, and GitHub Actions ran 45.4 billion jobs.
The report, released on October 28, 2026 on the GitHub blog, ties much of that growth to coding agents. In September 2026 more than 40% of pull requests began with an agent's first draft, up from 17% in March, and agents started 110 million pull requests in that month alone. GitHub also reports that 55% of GitHub Copilot monthly active users now use agentic features, up from 35% a year earlier, and that 61% of Copilot users who pick a model themselves now use more than one.
The platform has also kept growing in headcount. GitHub counts more than 225 million developers, with the United States at 38.3 million and India at 32.0 million, and both countries added more than 10 million developers during the year. TypeScript remains the most used language by contributors, ahead of Python and JavaScript, while Rust is the fastest-growing of the top languages in percentage terms.
··········
HOW GITHUB MEASURED THE 2026 OCTOVERSE YEAR.
The Octoverse year runs from October 1, 2025 to September 30, 2026 inclusive. This edition changes the method in ways that matter for anyone comparing it with older reports. Metrics now include both public and private repositories unless stated otherwise, while industry figures, open-source contribution counts and open-source project rankings remain limited to public activity. Earlier Octoverse editions covered public activity only and used a September 1 to August 31 window, so GitHub recalculated its 2024 and 2025 figures on the new definitions and window to keep the year-over-year comparisons consistent.
Several counting rules shape the totals. Each push counts once as a contribution, which is why commit totals can exceed contribution totals. GitHub Actions usage is counted in jobs rather than minutes. Contributor and contribution counts exclude accounts flagged as spam, GitHub staff and identifiable bots, and developers are assigned to countries through the location on their profiles, with a publication threshold of 100 developers per country. The exclusion of identifiable bots is relevant to the agent figures: activity started by agents is measured as a separate signal attached to pull requests, not as a population of bot accounts.
··········
PUSHES, COMMITS, PULL REQUESTS AND ACTIONS JOBS.
The core activity figures roughly tripled in a single year after two years of more moderate growth. The table below uses the values GitHub published for its three recalculated report years.
........
Metric | 2024 | 2025 | 2026 |
Pushes | 5.2 billion | 6.7 billion | 17.9 billion |
Commits | 10.0 billion | 12.9 billion | 34.3 billion |
Merged pull requests | 402 million | 538 million | 1.3 billion |
GitHub Actions jobs | 8.0 billion | 14.0 billion | 45.4 billion |
........
On a daily basis, 1.3 billion merged pull requests correspond to about 3.5 million merges per day, and 45.4 billion Actions jobs work out to roughly 124 million jobs per day, or about 1,440 per second, the rate GitHub itself quotes. Dependabot pull requests merged during the year passed 30 million, about double the 2025 pace, and open-source contributions reached 2.5 billion, more than double the 1.12 billion recorded in 2025.
The growth was not spread evenly across the twelve months. GitHub describes a steady rise from January followed by a surge around June 2026. Measured from September 2025 to September 2026, monthly commits grew about 6x, Actions jobs about 4.8x, pushes about 4.5x and merged pull requests about 3.9x. From June onward, about 55% of all commits went into repositories created that year, which suggests that a large share of the new volume came from fresh projects rather than from heavier activity in existing codebases.
The report also argues that the increase was not driven mainly by newcomers. Developers who pushed code in the past twelve months had been on GitHub for an average of 36 months, more than 75% of developers who opened a pull request in August pushed code again in September, and the average developer now works across 10% more repositories than a year earlier. GitHub highlights a few extreme individual cases, including a developer who made more than 1 million public contributions by using agentic workflows to maintain an API catalog.
··········
AGENTS NOW DRAFT MORE THAN 40% OF PULL REQUESTS.
The agent figures are the most consequential part of this year's report for teams evaluating coding tools. GitHub says more than 40% of pull requests opened in September 2026 started with an agent's first draft, compared with 17% in March. In absolute terms, agents started 110 million pull requests in September, about six times the March volume.
GitHub also compares the shape and outcome of agent-started work with pull requests where it detected no agent involvement. About 87% of agent-started pull requests change multiple files, against 79% for the other group. The merge rate within two weeks has moved in the agents' favour during the year: in February, 83% of agent-involved pull requests were merged within two weeks against about 86% of non-agent pull requests, while in September the figures were 90% for agent pull requests and about 86% for the rest. That is a measure of acceptance by reviewers and maintainers, not of code quality or defect rates, which the report does not quantify.
........
Agent and Copilot indicator | Figure reported by GitHub |
Pull requests started by an agent, March 2026 | 17% |
Pull requests started by an agent, September 2026 | More than 40% |
Agent-started pull requests in September 2026 | 110 million |
Agent pull requests merged within two weeks, September | 90% (about 86% for non-agent) |
Copilot monthly active users using agentic features | 55% (35% a year earlier) |
Copilot cloud agent usage growth | 230% year over year |
Copilot agent chat growth | 85% year over year |
........
Usage around the agents has grown as well. Stacked pull requests, launched as a preview on July 30, had been used by more than 264,000 developers by September. AI-assisted code review and GitHub Advanced Security usage each grew more than 90% year over year, and push protection, which blocks exposed secrets such as API keys and passwords before they land in a repository, ran 5.8 billion scans, nearly four times the 2025 pace. Public repositories mentioning interoperability or reverse engineering rose from about 30,000 in 2025 to more than 130,000 in 2026, which GitHub links to developers using agents to work with existing systems and file formats.
··········
MULTI-MODEL USE IN GITHUB COPILOT.
The report gives an unusually detailed view of how developers choose models inside GitHub Copilot. In September 2026, 61% of Copilot users who selected a model themselves used more than one, up from 50% four months earlier. Developers on paid plans used 29 distinct models in May and 47 in September, and no single model carried more than 25% of requests in September. Auto model selection, which lets Copilot choose the model, was used at least once that month by 63% of paying Copilot developers.
GitHub groups models into three families and reports how developer-selected requests split among them: 38 of every 100 requests went to general-purpose models, 37 to deep reasoning and planning models, and 25 to fast and lightweight models. The methodology lists examples for each group, with Claude Opus 5.5 and GPT 5.5 under deep reasoning and planning, Claude Sonnet 5, Grok 4.7 and GPT 6.1 Sol under general purpose, and Claude Haiku 5.5, Gemini 3.8 Flash and GPT 6 Luna under fast and lightweight. A model had to reach at least 1,000 weekly active users to be counted, and the trend data covers July 14 to October 13, 2026.
Three behavioural patterns stand out in GitHub's reading of the data. New models capture a larger share of requests than of developers in their first weeks, as early adopters use them intensively. Developers keep models in rotation after the novelty fades, so a model's share of developers declines more slowly than its share of requests. And few developers stay with a single model type for long, mixing reasoning, general and fast models within their work. The report connects these findings to Project HydraFusion, a research preview of model orchestration that routed almost 60,000 sessions in the Copilot CLI and the GitHub Copilot app and has since become generally available as GitHub Copilot Fusion.
··········
TYPESCRIPT, PYTHON AND RUST IN THE AGENT ERA.
TypeScript keeps first place by number of contributors, followed by Python and JavaScript. These three languages also recorded the largest gains in monthly contributors among the top ten between September 2025 and September 2026: 3.62 million more for TypeScript, 3.23 million for Python and 1.82 million for JavaScript. Rust is the fastest-growing of the top languages by percentage, having doubled during the year.
GitHub now reports a measure of AI involvement in merged pull requests by language, which shows that the languages growing fastest are also the ones where agents are most present.
........
Language | Growth in merged pull requests | AI involvement in merged pull requests |
TypeScript | 134% | 44.1% |
Python | 118% | 41.1% |
Rust | 162% | 40.4% |
Java | 30% | 21.2% |
........
C# and C++ fall between about 21% and 34% on these measures, and Java is the lowest on both. The repository data points the same way: among repositories created in 2026, TypeScript is the primary language in 35% of those with an agent signal, against 24% of those without one, and in older repositories the split is 27% against 17%. GitHub's forecasts for September 2027 put TypeScript at 13.07 million monthly contributors, Python at 12.56 million and JavaScript at 7.42 million, with C#, Java, Shell, PHP, Go, HCL and C++ ranging from 2.12 million down to 0.88 million. The association between agent use and typed languages is a correlation in GitHub's data; the report does not establish whether agents push developers toward TypeScript or whether TypeScript projects simply adopt agents earlier.
··········
225 MILLION DEVELOPERS AND THE COUNTRIES BEHIND THE GROWTH.
GitHub counts more than 225 million developers. The United States and India remain the largest communities and each added more than 10 million developers during the year, while Brazil added 3.28 million, the third-largest increase of any country.
........
Country | Developers on GitHub, September 2026 |
United States | 38.3 million |
India | 32.0 million |
China | 12.4 million |
Brazil | 10.2 million |
Japan | 7.15 million |
United Kingdom | 6.47 million |
Indonesia | 6.34 million |
Germany | 6.30 million |
Russia | 5.07 million |
France | 4.75 million |
........
Last year's Octoverse projected India at 57.5 million developers and the United States at 54.7 million by 2030. GitHub now reports that every country has closed at least 21.7% of the distance to its 2030 projection, with the United States at 38.6% and India at 28.3%. Morocco has progressed furthest at 54.4%, followed by Indonesia at 45.4%, while Kenya is lowest at 21.7%. The forecasting itself has become more elaborate: the 2031 outlook combines 43 time-series and regression models, against five in last year's report, and 21 of those models project growth of 3x or more. By pushes to public repositories, adjusted for profile location, the top ten countries are the United States, India, China, Brazil, Germany, the United Kingdom, Korea, France, Japan and Canada.
The industry analysis, based on public repositories classified into NAICS sectors by a fine-tuned multilingual embedding model, shows how unevenly AI mentions translate into agent use. In administrative and support services, mining, health care and agriculture, more than 40% of repositories mention AI, but agent-opened pull requests stay below 3%. Construction has the highest share of agent-opened pull requests of any sector, at about 9%, with about 25% of its repositories mentioning AI. The number of public repositories in the sample grew from 9.0 million in 2022 to 25 million in 2026.
··········
THE OPEN-SOURCE PROJECTS DRAWING CONTRIBUTORS.
The open-source rankings are dominated by AI tooling. Among single-vendor or product-led projects, the fastest-growing by contributors are Anthropic's claude-code, OpenAI's codex and Anomaly's opencode, followed by VS Code, GitHub's copilot-cli, Zed, Google's gemini-cli, BerriAI's LiteLLM, Microsoft's winget-pkgs and kilocode. Among community and independent projects, the list includes cc-switch, awesome-mcp-servers, Omarchy, first-contributions, llama.cpp, open-vsx.org, SGLang, vLLM, helium and vllm-ascend.
The most engaged community projects are OpenClaw, vLLM, llama.cpp, Omarchy, cc-switch, SGLang, PyTorch, Rust, Home Assistant core and CPython, while the most engaged product-led projects include Nous Research's hermes-agent, opencode, Earendil's pi, Stably AI's orca and Ping's t3code. Inference engines such as vLLM, SGLang and llama.cpp appearing in both the growth and engagement rankings indicates that local and self-hosted model serving remains one of the most active areas of open-source work alongside the coding agents themselves.
··········
WHAT THE NUMBERS MEAN FOR DEVELOPERS AND TEAMS USING GITHUB COPILOT.
For teams that build on GitHub, the report describes conditions that are already visible in day-to-day work. With agents drafting more than 40% of pull requests and changing multiple files in most of them, review capacity becomes the main constraint, and GitHub's own figures on AI-assisted code review and Advanced Security growing more than 90% show where organisations are directing effort. The higher two-week merge rate for agent pull requests in September indicates that reviewers are accepting more of this work, but teams still have to set their own standards for testing and ownership, since the report does not measure what happens to that code after merge.
The multi-model data is relevant to how Copilot is configured and paid for. With 47 models in use on paid plans, no model above a quarter of requests and 63% of paying developers trying auto model selection, standardising on one model is no longer how most Copilot users work. Organisations managing budgets will need to account for a mix of reasoning, general and fast models, and the routing approach behind Copilot Fusion is GitHub's answer to that mix. The surge in Actions jobs and push protection scans also signals more automated activity hitting CI pipelines and security checks, which is where agent-generated volume shows up in operating costs. Related changes in GitHub's agent tooling, such as GitHub Agentic Workflows and Copilot cloud sandboxes, are designed for this kind of continuous, automated workload.
··········
WHAT OCTOVERSE 2026 ESTABLISHES AND WHAT IT LEAVES OPEN.
The report establishes, on GitHub's own platform data, that activity tripled across pushes, commits, merged pull requests and Actions jobs in the year to September 2026, that the acceleration began around June, and that agents had become the starting point for more than 40% of pull requests by September. It also shows that model choice inside Copilot has fragmented across dozens of models and that agent use is concentrated in TypeScript, Python and Rust projects.
Several questions remain open. The inclusion of private repositories and the recalculation of earlier years make this edition comparable with itself but not directly with previous Octoverse reports. The agent signal is GitHub's own detection, and the report does not publish its error rate or describe agent activity that goes undetected. Merge rates measure acceptance rather than quality, and the report offers no data on defects, reverts or security incidents linked to agent-written code. The country and language forecasts are model projections rather than measurements, and GitHub has changed the forecasting method substantially since last year. Read with those limits in mind, Octoverse 2026 is the clearest platform-level evidence so far of how quickly coding agents have moved from experiment to routine part of software development.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
