Start here · Latency basics
What time to first token really measures
Time to first token, or TTFT, describes the interval between sending a request and receiving the first visible text.
Read the guide →Make sense of AI speed.
Practical explanations for the numbers, missing data and limitations behind performance claims.
Explore the essentials
Why tokens per second can mislead
Output throughput describes generation after a response starts.
Read guide →API performance is not ChatGPT app status
A native API request and a request made inside a consumer chat app are different journeys.
Read guide →A practical checklist when AI feels slow
Is the page failing to load, the first answer taking longer, or the response stopping midway? These symptoms need different evidence.
Read guide →What a probe region tells you
A probe region identifies where our measurement originates.
Read guide →Reading medians, p95 and sample counts
The median divides sorted observations into two halves.
Read guide →Reasoning time and visible answer speed
Some model configurations perform reasoning not emitted as ordinary answer text.
Read guide →Why a rate limit is not an outage
Rate limits can restrict requests or token use over time.
Read guide →When is there enough evidence for a best time?
Best-time recommendations need repeated observations across the compared hours and days.
Read guide →Prompt caching and fair speed comparisons
Monitoring repeats fixed prompts for comparability.
Read guide →What Run Live Test actually checks
Run Live Test times a bounded request from your browser to a small AI Fast Now object.
Read guide →How to read an AI performance report
A report should name the models, observation region, task profile, time window and sample count.
Read guide →