Which LLM API has the lowest time to first token?
Measured over the last 24 hours, the lowest median time to first token (TTFT) is 194 ms — typesafe running jev-1.13.0, requested from Europe (Germany). Read that as one measurement, not as a provider ranking: every provider here answers on a different model, and model size moves TTFT far more than the provider does. What is comparable like-for-like is the same provider on the same model from different regions — and that alone varies 4.7× for typesafe.
Measured time to first token, by region
| Requested from | Provider | Model answering | TTFT p50 | TTFT p95 | Samples |
|---|---|---|---|---|---|
| Europe (Germany) | typesafe | jev-1.13.0 | 194 ms | 229 ms | 72 |
| Europe (Germany) | groq | openai/gpt-oss-120b | 290 ms | 509 ms | 72 |
| South America (Santiago) | typesafe | jev-1.13.0 | 294 ms | 399 ms | 72 |
| US (Central) | typesafe | jev-1.13.0 | 597 ms | 812 ms | 71 |
| South America (Santiago) | groq | openai/gpt-oss-120b | 803 ms | 1032 ms | 72 |
| Asia (Tokyo) | typesafe | jev-1.13.0 | 909 ms | 1116 ms | 71 |
| Europe (Germany) | gemini-flash-lite-latest | 912 ms | 8150 ms | 72 | |
| US (Central) | groq | openai/gpt-oss-120b | 1307 ms | 1780 ms | 71 |
| Asia (Tokyo) | groq | openai/gpt-oss-120b | 1433 ms | 1860 ms | 71 |
| US (Central) | gemini-flash-lite-latest | 1943 ms | 8809 ms | 71 | |
| Asia (Tokyo) | gemini-flash-lite-latest | 2000 ms | 7613 ms | 71 | |
| South America (Santiago) | gemini-flash-lite-latest | 2006 ms | 7141 ms | 72 |
Real streamed completions, last 24 hours, requested from each probe region. The model is named on every row because it is the dominant factor in the number next to it.
What is the difference between time to first token and time to first byte?
Time to first byte (TTFB) is how long it takes the network and the provider’s edge to begin responding at all: the time from sending a request over an open TLS connection to the first byte back. DNS, TCP and TLS setup are timed separately. It does not involve the model. Time to first token (TTFT) is how long a real streamed completion takes to emit its first token, measured from opening a fresh connection: connection setup, the network round trip, queueing at the provider and the model’s prefill over your prompt.
TTFB measures endpoint response time. TTFT measures the wait until the first generated token. The two metrics are reported separately.
How much of time to first token is network latency?
Between 2% and 81% of time to first token is network time — the rest is the provider queuing your request and the model generating its first token. That is why a TTFT number quoted without the model name tells you almost nothing about the API: most of what it measures is the model.
| Requested from | Provider | TTFT p50 | TTFB p50 | Network share of TTFT |
|---|---|---|---|---|
| Europe (Germany) | typesafe | 194 ms | 157 ms | 81% |
| Europe (Germany) | groq | 290 ms | 218 ms | 75% |
| South America (Santiago) | typesafe | 294 ms | 174 ms | 59% |
| US (Central) | typesafe | 597 ms | 79 ms | 13% |
| South America (Santiago) | groq | 803 ms | 191 ms | 24% |
| Asia (Tokyo) | typesafe | 909 ms | 144 ms | 16% |
| Europe (Germany) | 912 ms | 93 ms | 10% | |
| US (Central) | groq | 1307 ms | 131 ms | 10% |
| Asia (Tokyo) | groq | 1433 ms | 155 ms | 11% |
| US (Central) | 1943 ms | 43 ms | 2% | |
| Asia (Tokyo) | 2000 ms | 66 ms | 3% | |
| South America (Santiago) | 2006 ms | 957 ms | 48% |
Can providers be ranked by time to first token?
These measurements use different models, named in the table above. Model size and inference behaviour affect TTFT, so these figures do not isolate provider performance. Use them to track each provider and model combination over time.
Measured from Europe (Germany), google has the lowest edge latency (93 ms TTFB) while typesafe has the lowest time to first token (194 ms TTFT, on jev-1.13.0). Both were measured from the same region on the same schedule, but the two metrics have different leaders.
Which providers have measured TTFT here, and why not all of them?
Inference is measured for 3 of 45 providers (google, groq, typesafe), because a streamed completion needs a paid API key for that provider and every probe spends tokens. Edge latency needs no key, so TTFB is published for all 45. We would rather publish 3 measured TTFT figures than estimate 45: an estimate presented as a measurement is the one thing this site does not do.
Method in full, including limits: methodology · Edge latency for every provider: fastest AI API by region · Availability: uptime comparison · Raw numbers under CC BY 4.0: JSON API.