Performance

Sova vs Axum vs Actix-web on an identical multi-page fixture site (same HTML/JSON bodies, verified SHA-256), including a realistic POST /api/echo JSON path.

Methodology

• Stand: bench/stand/ — shared fixtures, three release servers (stand_sova, stand_axum, stand_actix).
• Workspace [profile.release] uses thin LTO (codegen-units = 1) — production-shaped binaries, not dev.
• Bodies must match byte-for-byte across frameworks before load runs.
• Load tool: oha.
• Stability: oha warm-up 5s at full concurrency; 3 round(s) with rotating framework order; reported numbers are the median RPS (min/max kept in JSON for spread).
• This capture: profile deep, duration 15s, concurrency 100, TOKIO_WORKER_THREADS=4.
• Captured at 2026-08-09T06:19:10Z on Admins-MacBook-Pro.local.

Pages: /, /about, /blog, /blog/hello, /contact, /api/health, POST /api/echo.

Latest results — GET / (median)

Framework	Req/s	min–max	p50 (ms)	p99 (ms)
sova	122246	108062–123173	0.785	1.831
axum	122665	115021–124864	0.783	1.782
actix	100649	99291–104136	0.934	2.712

Latest results — POST /api/echo (median)

Framework	Req/s	p99 (ms)
sova	122813	2.106
axum	125580	1.947
actix	102865	3.013

Latest results — mean across all paths (of medians)

Framework	Mean Req/s	Mean p99 (ms)	max path spread %
sova	117589	2.272	12.4%
axum	120448	2.083	8.6%
actix	103513	3.017	7.4%

Per-path detail (median)

Framework	Path	Req Req/s	min–max	spread %	p50	p99
actix	/	100649	99291–104136	4.8%	0.934	2.712
actix	/about	104676	99866–105918	5.8%	0.929	2.751
actix	/api/echo	102865	102147–106977	4.7%	0.902	3.013
actix	/api/health	107683	100091–108013	7.4%	0.903	2.783
actix	/blog	99644	98465–103154	4.7%	0.925	3.448
actix	/blog/hello	104482	103196–106699	3.4%	0.908	3.900
actix	/contact	104591	100530–106134	5.4%	0.921	2.511
axum	/	122665	115021–124864	8.0%	0.783	1.782
axum	/about	123469	115999–125375	7.6%	0.781	1.740
axum	/api/echo	125580	117318–126586	7.4%	0.762	1.947
axum	/api/health	125295	123798–126208	1.9%	0.766	1.774
axum	/blog	115613	114235–122167	6.9%	0.779	2.220
axum	/blog/hello	117184	112162–122204	8.6%	0.782	2.384
axum	/contact	113332	113126–120328	6.4%	0.779	2.737
sova	/	122246	108062–123173	12.4%	0.785	1.831
sova	/about	115248	112862–115342	2.2%	0.786	2.502
sova	/api/echo	122813	120465–124537	3.3%	0.763	2.106
sova	/api/health	114522	113884–115791	1.7%	0.773	2.462
sova	/blog	118889	110540–119304	7.4%	0.786	2.180
sova	/blog/hello	110984	110878–114174	3.0%	0.792	2.712
sova	/contact	118417	109367–121292	10.1%	0.787	2.111

Re-run / regression gate

./bench/stand/run.sh                     # deep: 15s × 3 rounds, warm-up, median
PROFILE=quick ./bench/stand/run.sh      # smoke
./bench/stand/run.sh --update-baseline
ROUNDS=5 DURATION=20s ./bench/stand/run.sh

Regression thresholds (vs bench/stand/results/baseline.json): home RPS drop > 15% or p99 rise > 40% fails the script.

Machine-sensitive: compare relative rankings and deltas, not absolute RPS across laptops.