Skip to content
audiotale.net
Sign in
NewsNews

GitHub Copilot Fusion GA, Octoverse 2026: 34.3B commits, agents start 40% of PRs

GitHub made Copilot Fusion generally available at Universe and published Octoverse 2026, saying it added 4 million CPU cores to cope with AI agent load.

audiotale.net team

· 5 min read

Written by AI from the sources linked below. How we use AI

On this page
  1. Key points
  2. What to do
  3. What happened
  4. Background
  5. The numbers
  6. What it means
  7. What we don't know yet
  8. What to watch next
  9. Sources

GitHub used its Universe conference on 28 October 2026 to make Copilot Fusion generally available, give GitHub Enterprise customers access to Copilot without a separate subscription, and publish its Octoverse report, which counts 34.3 billion commits in the 12 months to 30 September 2026. The same day, GitHub's EVP of CoreAI Jay Parikh said the company has added 4 million CPU cores and 160 petabytes of storage this year to keep up with AI agents that now start more than 40% of pull requests.

Key points

  • Copilot Fusion, which picks and combines models for each task, is generally available on all Copilot plans in VS Code, the GitHub Copilot app, Copilot CLI and JetBrains IDEs.
  • Octoverse 2026 counts 34.3 billion commits, 17.9 billion pushes and 1.3 billion merged pull requests in the 12 months ending 30 September 2026, with more than 225 million developers.
  • Agents started more than 40% of pull requests in September 2026, up from 17% in March, according to GitHub.
  • GitHub says it added 4 million CPU cores, 160 PB of storage and 50 Tb/s of network capacity in 2026, and expects to finish a multi-region Azure move early next year.
  • GitHub Enterprise customers get all Copilot features without a separate subscription and pay for AI credits used.

What to do

If you manage a GitHub organisation, check three things this week. First, Fusion only uses models your enterprise and organisation policies allow, so review the model policy before turning it on, since admins can disable individual models for Fusion without disabling them for everyone. Second, Enterprise admins can set Fusion at any effort level as the default model through Enterprise managed settings. Third, because billing is per model used, with a limited-time 30% discount for paid plans, watch spending during the first weeks. To try it: update your client, open the model picker, choose Copilot Fusion, then pick Standard, High or Max.

What happened

GitHub's Universe post, written by Jay Parikh, ties the day's announcements to load. The company says it has added 4 million CPU cores, 160 PB of high-speed storage and 50 Tb/s of network capacity so far in 2026, and is moving to a multi-region deployment in Azure, expected to complete early next year. The Register and The New Stack, both citing GitHub, report 7.4 billion commits in September 2026 and describe a roughly fivefold rise on a year earlier. GitHub's own Octoverse page expresses the monthly trend as an index (September 2026 commits at about six times September 2025), so the exact multiple differs between sources; we use only the figures GitHub published directly in its report.

On the product side, Copilot Fusion combines the multi-model orchestration of the research preview called Project HydraFusion with Copilot's auto model selection. You pick Fusion instead of one model and Copilot chooses. It runs three workflows: Single (one model does the task), Cascade (an efficient model drafts and a quality check either accepts the result or escalates to a stronger one) and Critique (one model drafts, another reviews and the first revises). Preview users of auto, HydraFusion or HydraFusion Max move automatically to the matching Fusion levels.

GitHub says Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the average cost per task and 40% less time per task. That is a vendor claim citing a third-party index; we have not seen the underlying run.

Background

Octoverse is GitHub's annual report. Its year runs from 1 October to 30 September. Reliability has been a theme: The Register reports that AI-driven commit volume strained the platform from December, with problems in February, April and August, and a 90-day availability figure that reportedly fell to about 84% on 7 May, based on reconstructed status-page data rather than GitHub's own reporting. We treat that low as unconfirmed. GitHub's post links a September 2026 availability report that it says records two incidents with degraded performance.

For earlier coverage of how the tools around Copilot are changing, see our note on Microsoft's Pylance moving to the Pyrefly type server.

The numbers

Octoverse compares each year with the two before. Commits more than tripled against 2024.

Commits on GitHub, 12 months to 30 Septemberbillion
Commits on GitHub, 12 months to 30 September (billion)
billion
202410.0
202512.9
202634.3

Source: GitHub Octoverse 2026

Merged pull requests rose from 402 million (2024) to 538 million (2025) and 1.3 billion (2026), about 3.5 million a day. GitHub Actions ran 45.4 billion jobs, against 14.0 billion a year earlier and 8.0 billion in 2024.

Share of pull requests started by agents%
Share of pull requests started by agents (%)
%
Mar 202617
Sep 202640

Source: GitHub Octoverse 2026

The September bar shows "more than 40%", so 40 is a floor. GitHub counts 110 million agent-started pull requests in September, about six times March. It also says 83% of agent pull requests were merged within two weeks in February, rising to 90% in September, while a comparison group stayed near 86%.

Developer counts: more than 225 million. The largest communities are the United States (38.3 million), India (32.0 million), China (12.4 million), Brazil (10.2 million) and Japan (7.15 million).

On models, 61% of Copilot users who choose a model used more than one in September (50% four months earlier). Paying users used 47 distinct models in September, up from 29 in May, and the largest single model accounted for 25% of requests or less. That is the case for routing, and the reason Fusion exists.

What it means

For developers, Fusion makes model choice a default rather than a daily decision. The Cascade and Critique workflows trade extra calls for quality or cost, and since billing is per model used, a Critique run costs more than a Single one. The 30% discount softens that for now, but it is time-limited.

For companies, folding Copilot into GitHub Enterprise removes a procurement step, but the cost moves from seats to credits. Heavy agent use means heavy token bills. GitHub gives no prices in the post.

For the platform, the numbers say why reliability matters. An agent can open many pull requests in the time a person opens one, and Octoverse notes that about 55% of commits from June onward came from brand-new repositories. Expect the load to keep growing.

For security, GitHub also announced reachability analysis for Dependabot alerts and a security extension for the Copilot app, per its changelog headlines; we did not review details.

What we don't know yet

  • The pricing of the Enterprise Copilot credits, and when self-serve purchase arrives ("coming soon").
  • Independent evidence for Fusion's cost and speed claims.
  • The exact September commit count and growth multiple: sources differ between five and six times a year earlier.
  • Whether the Azure migration, due early next year, ends the reliability problems.
  • Which models Fusion uses at each effort level; GitHub's page does not list them.

What to watch next

Watch GitHub's monthly availability reports, the end of the 30% Fusion discount, which additional clients get Fusion, and whether other IDE vendors copy the Cascade and Critique pattern.

Sources

SaveStart the discussion

Written by

audiotale.net team

The official account of the people who run audiotale.net: announcements, conversation starters and the newsroom's drafts. Posts written with AI say so. Every other account here is a real person.

Read next