Programming Language &
Database Benchmark Results

October 10, 2026 — one standardized campaign comparing API implementations across PostgreSQL, MariaDB, Cassandra, YugabyteDB, and MongoDB.

Both limited-resource and full-resource tests are shown. Failed tests and connection preflight failures remain part of the published record.

October 10 Campaign

Loading the original benchmark results…

Download the original raw results (TSV)

Source: five_database_campaign_20261010_011548/summary.tsv. This public file is read directly by JavaScript to build the tables below. It contains the original campaign summary, not hand-entered figures.

Historical results are preserved. The previous benchmark page is available at October 9, 2026 historical results. Earlier tests are not mixed into the October 10 campaign.

Jump to a table

Highest Error-Free Throughput

For each database and resource mode, this shows the highest reported requests/sec at a tested concurrency level with zero recorded errors. These are individual peak measurements, not overall rankings for every workload.

DatabaseResource modeLanguageConcurrencyRequests/sec
Loading results…

Complete Results — All Ten Tables

Every table displays seven concurrency levels: C1, C10, C50, C100, C250, C500, and C1000. Each cell gives the recorded requests/sec plus the number of errors. Click a column heading to sort; choose the metric above each table.

Preflight failed means the API did not pass its initial check, so no load measurements were taken at any concurrency level. It is not a result of 0 requests/sec. All four preflight failures in this campaign were Zig with PostgreSQL or YugabyteDB, in both resource modes. One Ruby/PostgreSQL failure has an error count recorded as unknown in the original summary; it is displayed as unknown, not zero.

Test Methodology and Interpretation

Every API/database combination was tested independently, with one API and its target database running at a time. Each concurrency level ran for approximately 30 seconds, with an interval between levels and database stabilization before the next group.

Limited API2 CPU
2 GB RAM
Limited database4 CPU
8 GB RAM
Full API12 CPU
90 GB RAM
Full database12 CPU
90 GB RAM

The concurrency values represent simultaneous simulated clients. An individual high-throughput measurement is not necessarily the most reliable production choice. Error counts must be considered alongside throughput and latency.

The official raw TSV contains mode, database, language, concurrency, request count, requests/sec, errors and status. Detailed latency measurements remain in the benchmark server's individual run logs, not in this summary file.

No results have been repaired, replaced, or silently excluded for this publication. Zig preflight failures remain explicitly visible; targeted future testing can be published as a separate dated run.