QWED historical latency measurements (v1.1.0)

Historical record, not current figures. These numbers were measured on QWED v1.1.0 in December 2025. The engines, their checks and the fail-closed behaviour have changed a great deal since then, so latency on the current release will differ. Measure on your own workload and hardware before relying on any figure. QWED’s current evaluation results are at qwedai.tech/evidence.Cost comparisons, hosted pricing, an ROI calculator and comparisons with other tools previously on this page were removed: none of them had a reproducible source.

Test environment (December 2025):

Hardware: AWS EC2 t3.medium (2 vCPU, 4GB RAM)
Python: 3.11
QWED: v1.1.0
Date: December 2025

Latency (v1.1.0)

Verification engine latency

Engine	Operation	Typical Latency	P95 Latency	P99 Latency
Math	Simple arithmetic	5ms	12ms	18ms
Math	Calculus (derivative)	15ms	28ms	45ms
Math	Matrix (3×3)	22ms	40ms	65ms
Math	Financial (compound interest)	8ms	15ms	25ms
Logic	Simple SAT	25ms	50ms	80ms
Logic	Z3 theorem proving	120ms	250ms	400ms
Code	AST security scan	35ms	70ms	110ms
Code	Symbolic execution (simple)	2,500ms	8,000ms	15,000ms
SQL	Query parsing	18ms	35ms	55ms
SQL	Schema validation	45ms	85ms	130ms
Stats	Mean/median	12ms	25ms	40ms
Stats	Regression (100 rows)	95ms	180ms	280ms
Fact	TF-IDF grounding	60ms	120ms	200ms
Image	Metadata check	15ms	30ms	50ms
Consensus	3-model check	4,500ms	7,000ms	12,000ms

Observations at the time

Most single-engine checks finished in under 100ms on this hardware.
⚠️ Symbolic execution was slow: use it only for simple functions.
⚠️ Consensus was expensive: it makes several LLM API calls.

Latency optimization tips

1. Enable Redis caching

from qwed_sdk import QWEDClient

client = QWEDClient(
    cache_enabled=True,
    redis_url="redis://localhost:6379"
)

# Repeated verifications are served from the cache

2. Async verification

import asyncio

async def verify_batch(items):
    tasks = [client.verify_async(item) for item in items]
    return await asyncio.gather(*tasks)

# Runs independent verifications concurrently

3. Selective engines

# Only enable engines you need
client = QWEDClient(engines=["math", "code"])  # Faster startup

4. Timeout tuning

# Reduce timeout for simple operations
result = engine.verify_math(expr, timeout_ms=1000)  # 1s max

Throughput (v1.1.0)

Configuration	Requests/Second	Avg Latency
Single thread	15 req/s	65ms
4 workers	55 req/s	70ms
8 workers	95 req/s	80ms
16 workers (with Redis)	180 req/s	85ms

These figures came from the same December 2025 environment. Size your own deployment from your own load tests.

Need help measuring? Contact: support@qwedai.com

Architecture

Concepts

Specifications

Integration Patterns

Contributing

QWED historical latency measurements (v1.1.0)

Latency (v1.1.0)

Verification engine latency

Observations at the time

Latency optimization tips

1. Enable Redis caching

2. Async verification

3. Selective engines

4. Timeout tuning

Throughput (v1.1.0)

​Latency (v1.1.0)

​Verification engine latency

​Observations at the time

​Latency optimization tips

​1. Enable Redis caching

​2. Async verification

​3. Selective engines

​4. Timeout tuning

​Throughput (v1.1.0)

Latency (v1.1.0)

Verification engine latency

Observations at the time

Latency optimization tips

1. Enable Redis caching

2. Async verification

3. Selective engines

4. Timeout tuning

Throughput (v1.1.0)