On Oct 28 GitHub published its Octoverse 2026 report: developers pushed code 17.9 billion times in the past 12 months, more than 2024 and 2025 combined, with 34.3 billion commits and 1.3 billion merged pull requests (3.5 million a day). By September more than 40% of pull requests started with an agent draft, up from 17% in March; 55% of Copilot monthly active users use agentic features, and paid Copilot users used 47 distinct models in September.
Key Takeaways
- ✓17.9 billion pushes, 34.3 billion commits and 1.3 billion merged pull requests in the past 12 months, about 3.5 million merges a day (GitHub figures)
- ✓More than 40% of pull requests started with an agent draft by September 2026, up from 17% in March; that was 110 million in September alone, about 6x March
- ✓55% of Copilot monthly active users now use agentic features, up from 35% a year ago, and Copilot cloud agent grew 230% year over year
- ✓Paid Copilot users used 29 distinct models in May and 47 in September, with no model above 25% of requests; 61% of users who pick models mix more than one
- ✓GitHub now serves more than 225 million developers; the US has 38.3 million and India 32 million, each adding over 10 million this year
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and pain point
Octoverse is GitHub's annual platform data report, and this year's theme is in the title: developers are building with agents. Coding agents went from experiments to daily tools over the past year, yet the industry has lacked large-scale usage data; most claims come from vendor demos or small surveys. This report draws on GitHub's own telemetry across more than 225 million developers, one of the largest first-party datasets available. All figures are GitHub's own, and GitHub also sells Copilot.
Architecture and mechanism
GitHub says activity climbed from January and surged around June, as several shifts landed at once: stronger models, maturing harnesses around them, and agents moving from single sessions to persistent ones that track progress across days, orchestrate other specialized agents, and run heavily in the cloud. Among repos created this year, the average repo with an agent configuration file logged eight times as many commits as one without. For the first time GitHub used a language model to classify sampled public repos by NAICS industry, tracking whether a repo mentions AI and whether an agent opened a pull request; construction leads with about 9% of repos having agent-opened PRs.
Benchmarks and comparisons
These are usage figures, not model benchmarks (GitHub figures): 17.9 billion pushes, 34.3 billion commits (more than triple two years ago) and 1.3 billion merged PRs. More than 40% of PRs started with an agent draft in September, versus 17% in March. Agent PRs merged within two weeks 90% of the time in September, against about 86% for others. 55% of Copilot monthly actives use agentic features; cloud agent grew 230% and agent chat 85% year over year. Paid users used 47 models in September, none above 25% of requests. Actions ran 1,440 jobs per second and push protection ran 5.8 billion scans.
What developers can do now
The clearest signal is that agent-drafts-first, human-reviews-second is becoming the default workflow, so review skills matter more than typing code. Add an agent configuration file to your repo (such as AGENTS.md or Copilot instructions), hand smaller scoped tasks to the cloud agent, and turn on AI code review and push protection. On models, no single one dominates; switching between reasoning, general and lightweight models per task is common. Full data is on the GitHub Blog and octoverse.github.com.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.