WORLDTECHOpenAI's Pando backend reportedly speeds Git pushes by about 80%
According to a report by Crypto Briefing, OpenAI has introduced Pando, a Git-compatible backend that reportedly cuts the time needed to push code to its monorepo by about 80 percent. The report presents the figure as coming from OpenAI's own research findings.
What the backend does for developers
Pando now serves all development in OpenAI's monorepo, and the reporting says median push latency fell by nearly 80 percent, a drop that works out to almost a fivefold improvement.
The backend keeps the standard Git client and protocol, so engineers and Codex agents continue to run git push and git fetch in the usual way. Pando treats branches independently, permitting parallel updates to unrelated branches while preserving the order of changes within each branch.
Scaling pressure that prompted the project
OpenAI says its repository was under strain from a flood of commits produced by Codex, its AI coding agent. Pando addresses two separate scaling problems: many writers targeting different branches at once, and heavy read traffic from fetches and clones.
According to the research findings, the system entered production in April 2026 after roughly ten weeks of development. One engineer led the work, with substantial assistance from teams of Codex agents.
Reported throughput figures
The reporting says Pando serves close to one petabyte of Git data each day while running on only a handful of CPU cores. At its recorded peak it pushed more than 57 TB of Git pack data within a single hour.
The research notes that clone and fetch operations can draw up to 10 Gbps of bandwidth on standard cloud virtual machines. OpenAI has also contributed work back to the Git project itself.
Open questions
- How were the latency measurements across OpenAI's monorepo carried out?
- Will OpenAI publish independent benchmarks of Pando?