If you watch the FHIR tooling space, the small newsworthy thing this week is not the leaderboard on a new benchmark. It is what happened a few hours after the leaderboard went live. Health Samurai published an open-source FHIR server benchmark on June 29, 2026, comparing Aidbox, HAPI FHIR, Medplum, and the Microsoft FHIR Server. Almost immediately, the Medplum CTO publicly forked the repository on GitHub. That fork is the part worth reading.
Worth saying once: the benchmark is vendor-run, since Health Samurai builds Aidbox. The fork is interesting precisely because it is how an open benchmark is supposed to work.
What a Public Fork Actually Signals
A public fork from a competing vendor's CTO is not a panic move. It is the opposite, frankly. It says the harness is reproducible enough to be worth running locally, and that there is enough public attention on the result that an optimisation pass makes sense. That is healthier than the closed-door pattern where a vendor disputes a number in a blog post and nobody else can check.
For more FHIR background reading, the wider site covers similar reproducibility questions across the FHIR stack.
Where Medplum Currently Sits on the Numbers
The 2026-06-29 snapshot has Medplum at about 1,420 CRUD requests per second, 764 resources per second on bundle import, and 1,796 RPS on FHIR search. The search number is the one most people skim past, and it is the most interesting line for Medplum. At 1,796 RPS on search, Medplum is second across the four servers tested, ahead of HAPI's 1,005 RPS. That is a real strength, not a weakness to be defended.
The bundle import number is where the optimisation work probably starts. Aidbox is at 2,678 res/sec on the same workload and HAPI is at 2,214, which puts Medplum's 764 in a gap the team has room to close. The Medplum benchmark notes also flag that composite search is currently unsupported, which is the obvious spec-completeness item to ship.
What the Fork Probably Does Next
Forking the harness gives the Medplum team three useful things. They can rerun the suite privately against unreleased builds, they can submit pull requests back to the upstream repo with reproducible improvements, and they can publish their own numbers without anyone questioning whether the workload was the same. All three of those are open-source-ecosystem-positive moves.
If you want to look at what was forked, the public benchmark repo holds the harness, the k6 scripts, the per-container resource limits, and the docker-compose definitions. The bundle vs individual posts comparison covers the workload shape that bundle import is testing, which is the metric most likely to move first.
Why This Is a Healthy Dynamic
The most useful framing here is that open benchmarks change the incentive shape for every vendor in the comparison. Aidbox, HAPI, Medplum, and the Microsoft FHIR Server all have a reproducible target to optimise against, and the daily-rerun dashboard makes every shift visible. The Medplum CTO forking the repo is the first concrete sign that the dynamic works.
It is the same pattern that benefits every open-source ecosystem: visibility, reproducibility, and a public scoreboard. The HAPI vs Ontoserver comparison for mid-size health systems goes through a similar two-tool dynamic in the terminology space, and the lessons port across.
The Real Outcome to Watch
In three to six months, the more telling signal will be whether the other servers in the comparison also fork, contribute, or post their own optimisation work. If they do, this benchmark stops being a snapshot and becomes a living thing. That is the part of the news that matters more than today's leaderboard.