State of API Performance

Latency benchmarks, error rate patterns, and performance insights across web frameworks.

Last updated: March 2026

Frameworks benchmarked

Languages covered

47ms

Median P50 latency

2.3%

Avg error rate

API Latency by Framework

Median (P50), P95, and P99 response times for common web frameworks handling typical CRUD API workloads.

Gin (Go)

8ms

Fastify (Node.js)

12ms

ASP.NET Core

15ms

Express (Node.js)

18ms

Spring Boot (Java)

22ms

FastAPI (Python)

25ms

Laravel (PHP)

35ms

Django (Python)

38ms

Rails (Ruby)

42ms

Key Takeaways

Gin (Go) and Fastify (Node.js) lead with sub-10ms P50 latencies
P99 latency is 3-8x higher than P50 across all frameworks
Django and Rails show the widest P50-to-P99 gap due to ORM query variance

Error Rates by Framework

Percentage of API requests resulting in 5xx errors, averaged across production applications.

Gin (Go)

1.2%

ASP.NET Core

1.5%

Spring Boot

1.8%

Fastify

FastAPI

2.1%

Express

2.4%

Django

2.8%

Rails

Laravel

3.2%

Top Performance Bottlenecks

The most common causes of high latency identified in distributed traces, ranked by frequency.

Rank	Bottleneck	Frequency	Impact
#1	Database queries Slow queries, N+1 patterns, missing indexes	68%	High
#2	External API calls Third-party service timeouts and slow responses	45%	High
#3	Serialization overhead JSON marshaling of large payloads	32%	Medium
#4	Authentication middleware Token validation, session lookup, RBAC checks	28%	Medium
#5	Connection pool exhaustion DB or Redis pool saturation under load	22%	High
#6	Cold starts Serverless and container startup latency	18%	Medium
#7	Garbage collection pauses GC-heavy languages under memory pressure	15%	Low

Database Query Latency Patterns

Average latency contribution of database operations in traced API requests.

Time Spent in DB (pct of total request)

Simple CRUD25%

Joins (2-3 tables)40%

Complex queries55%

N+1 patterns72%

N+1 Query Detection

34%

of traced applications have N+1 query patterns

4.2x

average speedup after fixing N+1 queries

Methodology

This report aggregates anonymized performance data from production applications monitored by TraceKit, supplemented with publicly available benchmark data from framework maintainers and the OpenTelemetry community. All data is anonymized and aggregated.

See where your APIs stand

Get latency percentiles, error rates, and bottleneck detection for your production APIs.

Start free Try the Latency Calculator

Related Resources

Framework Observability

Setup guides for Gin, Express, Django, Spring Boot, and more

Trace Visualizer

Paste and visualize your trace data as a waterfall or flame graph

Distributed Tracing Guide

Complete guide to distributed tracing with OpenTelemetry

Debug Latency Issues

How to debug latency issues without redeploying