The 2026 RTOS Performance Report benchmarks eight kernels (FreeRTOS, Eclipse ThreadX, PX5 RTOS, Zephyr, Arm Keil RTX5, RT-Thread, NuttX, Micrium uC/OS-III) across ten tests on an STM32L475VG Cortex-M4F board at 80 MHz using GCC 13.3. New additions over the 2024 edition include the full Thread-Metric suite with interrupt tests, an oscilloscope-measured interrupt-latency study, a POSIX API comparison, an IAR vs. GCC compiler study, and cross-architecture data on Cortex-M33 boards. Two key findings: compiler optimization level (-O0 to -O2) affects throughput by 10–50% and is a bigger lever than kernel choice; dispatch latency (not IRQ entry) spans 7.6x at p99 across kernels and is the real differentiator for interrupt-driven designs. Full per-kernel rankings and latency tables are in the downloadable report.
Table of contents
What the 2026 RTOS Benchmark Study MeasuresWhat’s New in the 2026 Edition of the RTOS Benchmark StudyWhy the RTOS Benchmark Study Methodology Is the StoryTwo RTOS Benchmark Study Findings You Can Use Right NowWhat’s in the ReportUse the RTOS Benchmark Study Data to Narrow the FieldQuestions this post answers
What is the difference between IRQ latency and dispatch latency in RTOS benchmarking?
IRQ latency is the time from a hardware interrupt firing to the first instruction of the interrupt handler — on Cortex-M at 80 MHz, most kernels cluster tightly here because it is dominated by processor exception entry. Dispatch latency is the time from the handler signaling a higher-priority thread to that thread actually running. Dispatch latency spans 7.6x at the 99th percentile across kernels, making it the real differentiator for interrupt-driven designs. Engineers sizing up RTOS options for deadline-sensitive designs track findings like these on daily.dev.
How much does compiler optimization level affect RTOS benchmark throughput on Cortex-M?
Moving from -O0 to -O2 lifts kernel-test throughput by roughly 10 to 50%, and the bare calibration loop runs about 2.75x faster — same C code, same silicon, same kernel. At -O0 the calibration spread reaches 3.6%, failing the study's own 1% repeatability rule; at -O2 and above it closes to 0.9%. Build settings are a first-order performance lever, largely independent of which RTOS is chosen. Developers optimizing embedded firmware builds find compiler trade-off data like this on daily.dev.
Which RTOSes were added in the 2026 RTOS benchmark study compared to the 2024 edition?
The 2024 study covered FreeRTOS, Eclipse ThreadX, PX5 RTOS, and Zephyr across five tests. The 2026 edition adds Arm Keil RTX5, RT-Thread, NuttX, and Micrium uC/OS-III, bringing the total to eight kernels across ten tests. The expanded cohort covers commercial, foundation-backed, and community open-source kernels, and a year-over-year chapter tracks how the original four kernels changed between editions. Teams narrowing an RTOS shortlist for a new product keep up with benchmark data like this on daily.dev.