What We Measure
MyAISpeed measures how fast AI providers respond from your actual location, on your actual network. Unlike benchmarks that test from controlled data centers, our measurements reflect the real-world experience of real users.
Metrics Definitions
- TTFB (Time to First Byte)
- The time from when a request is sent to when the first byte of the response is received. This measures the combined latency of DNS resolution, TCP/TLS handshake, network transit, and server processing. Lower is better.
- TTFT (Time to First Token)
- The time from when an inference request is sent to when the AI model produces its first output token. This includes TTFB plus the model's processing/queue time. Available when users provide their own API key. Lower is better.
- TPS (Tokens Per Second)
- The rate at which the AI model generates output tokens after the first token. This measures raw inference throughput. Higher is better.
- Success Rate
- The percentage of test requests that complete successfully without errors or timeouts. Higher is better.
Test Types
Connection Test (Default)
Every user can run a connection test without any API key. We send a lightweight request to each AI provider's API endpoint and measure the round-trip time. This test runs through our edge proxy (Cloudflare Workers, deployed across 300+ global locations) to ensure the measurement reflects geographic routing quality.
Inference Test (Optional, BYOK)
Users who provide their own API keys can run a full inference test. We send a standardized prompt (Count from 1 to 20) to each provider and measure TTFT and TPS using streaming responses. The prompt is intentionally simple and short to isolate network and inference performance from model reasoning complexity.
Data Collection
Each test automatically collects anonymized metadata:
- Geographic location (country, region, city) derived from IP via Cloudflare's edge network
- Network provider (ISP/ASN) from Cloudflare's routing data
- Device type (mobile, desktop, tablet) from User-Agent
- Connection type (WiFi, 4G, 5G) from the Network Information API when available
- Browser and OS from User-Agent
We do not collect personally identifiable information. IP addresses are never stored. All geographic and network data is derived from Cloudflare's edge infrastructure and stored in aggregate form only.
API Key Security
When users provide API keys for inference testing:
- Keys are stored only in the user's browser (localStorage)
- Keys are transmitted over HTTPS to our edge proxy
- Keys are used for a single request and immediately discarded in memory
- Keys are never logged, stored in any database, or included in any response
- Our Worker code explicitly nullifies key variables after use
Rate Limiting
To ensure data quality and prevent abuse, tests are rate-limited to 3 per minute and 10 per hour per IP address. IP addresses are hashed with a salt before rate limit checks and are never stored in plaintext.
AI Speed Index
The AI Speed Index aggregates data from all user tests over a rolling 30-day window. Rankings are based on average TTFB across all successful tests. Country-level breakdowns are available when sufficient data exists for a given region.
Limitations
- Connection tests measure edge-to-provider latency through our proxy, which adds a small amount of overhead (typically <5ms) from the Cloudflare edge
- Inference test results depend on the user's API tier and rate limits with each provider
- Results may vary based on time of day, provider load, and network conditions
- The Network Information API is not available in all browsers, so connection type data may be incomplete
Open Data
We believe in transparency. Aggregated, anonymized data from the AI Speed Index is available through our public API. If you're a researcher, journalist, or developer interested in AI performance data, reach out.