THE QUICK ANSWER
Pando is OpenAI’s internal Git backend. It preserves Git workflows while scaling concurrent repository traffic.
What OpenAI announced
OpenAI reports roughly 80% lower median push latency and nearly one petabyte of Git data served daily. These measurements describe its own monorepo.
- Writes: unrelated branches can advance independently.
- Reads: reusable packfiles and regional caches reduce repeated server computation.
- Tradeoffs: cross-branch reads need not represent one instant; clients may receive redundant objects.
Pando and Scion: different boundaries
Anthropic’s Scion centers on online clients and server-side commits, merges and history queries. Its clients keep bounded caches and request working-copy files on demand.
| Design question | OpenAI Pando | Anthropic Scion |
|---|---|---|
| Where is the change? | Git serving backend | Source-control system and client |
| How is work shared? | Git commits and pushes | Server-created changesets |
| How are reads scaled? | Reusable packs | Object caches and sparse working copies |
Can you use it today?
The Pando announcement provides no public backend installer or service signup. OpenAI says it is exploring wider access; its public Git improvements are separate from Pando itself.
Our take: measure the queue first
For a team evaluating infrastructure, start with the slow operation: workspace preparation, publishing a change, CI startup, or review. Record a representative workload before selecting a remedy. A strong result at another company does not identify your bottleneck.
Use our CI checkout calculator to estimate runner time spent preparing repositories. Treat that estimate as a baseline for a controlled experiment, not a prediction of Pando savings.
For more context, read the GitHub and Scion comparison. Keep architectural choices, public availability and measured outcomes as separate evaluation questions.
Common questions
Does Pando replace Git?
It changes the backend while retaining Git compatibility.
Is this a Pando installation guide?
No. This is a sourced architecture brief; no deployment or benchmark was performed.
Which system is faster: Pando or Scion?
These announcements do not provide a shared benchmark. Ranking them would require equivalent repositories, operations, concurrency and hardware.
Sources & verification
Checked on 3 November 2026 against the official engineering articles. OpenAI published its Pando account on 2 November. Performance figures are vendor reports; we have not tested Pando. The linked X post prompted this brief; its page could not be fetched directly.
- OpenAI: Scaling Git at OpenAI — 2 November 2026
- OpenAIDevs: original Pando announcement on X
- Anthropic: Redesigning source control
Written with AI assistance and checked against linked primary sources. This site has not benchmarked Anthropic Scion. Read our editorial policy.