For US health systems that have spent years asking FHIR vendors for trustworthy performance numbers, something small but real just happened. Health Samurai released an open-source FHIR server performance benchmark on June 29, 2026, comparing Aidbox, HAPI FHIR, Medplum, and the Microsoft FHIR Server on identical hardware. Within hours, the Medplum CTO publicly forked the repository on GitHub. That fork is the part of the story that matters more than the leaderboard.
Worth saying once: Health Samurai builds Aidbox, so the benchmark is vendor-run. The shape of the response is what makes it interesting. Open benchmarks invite vendor pushback, and vendor pushback through an open repo is healthier than the alternative.
What a Public Fork Actually Means in This Market
When a competing vendor's CTO forks an open benchmark, the message is not defensive. It is that the harness is credible enough to optimise against, and that the result is going to be referenced enough in customer conversations that an engineering response is worth shipping. For US hospital procurement teams, that is the most important byproduct of any vendor benchmark: a competitive dynamic that pushes every server in the comparison to publish better numbers, not worse explanations.
You can see the harness for yourself in the open-source repository. The Medplum fork sits alongside it on GitHub. For broader interoperability context, the US healthcare interoperability hub collects the related coverage.
The Medplum Numbers Worth Looking At First
The fork makes more sense when you look at where Medplum currently lands in the 2026-06-29 snapshot. CRUD throughput is at about 1,420 requests per second. Bundle import is at 764 resources per second. FHIR search is at 1,796 RPS, which is the most interesting line in the table for Medplum. At 1,796 RPS on search, Medplum is second across the four servers tested, ahead of HAPI's 1,005. That is a real strength, not a gap.
Where the fork probably starts optimising is the composite-search support. The benchmark notes that composite search is currently unsupported on Medplum, which is the spec-completeness item most likely to ship next. Bundle import is the second target, since the 764 res/sec number sits well below Aidbox's 2,678 and HAPI's 2,214 on the same workload. For US health systems doing legacy EHR migrations, bundle import is the metric that decides whether the cutover weekend works.
Why This Pattern Is Good for American Health Systems
US health systems that procure FHIR servers have been arguing with vendor decks for years. The shift here is that the decks now have to point at a public repo and a daily-rerun dashboard. That is a different kind of evidence, and it is the same shift that benefits American buyers in any other open-tooling market.
The honest framing for hospital IT teams is that this benchmark does not pick a winner. It picks a baseline that every vendor in the comparison will now spend engineering hours moving. The HAPI vs Aidbox vs Medplum comparison for US health systems covers the broader procurement framing those three servers usually sit inside.
What to Watch Over the Next Six Months
Three signals are worth tracking. First, whether Medplum ships composite search and posts a bundle-import optimisation. The fork makes that work visible. Second, whether HAPI or the Microsoft FHIR Server respond in kind, either by contributing changes upstream or by publishing their own re-runs. Third, whether Health Samurai keeps the daily-rerun cadence honest across vendor responses, since the value of the benchmark is the rerun, not the snapshot.
For US health systems already in active FHIR procurement, the practical effect is small in the short term and meaningful in the medium term. The open-source vs commercial terminology servers for US hospitals writeup goes through a parallel open-vs-closed evaluation pattern in the terminology space, and the same dynamic now applies to the FHIR server tier.
The Real Takeaway
The interesting thing about this fork is not that Medplum is now optimising against a Health Samurai harness. It is that a public, reproducible FHIR server benchmark with a vendor response is finally a normal thing to point at in 2026.
Related Posts
