What first-token latency measures
First-token latency measures how long users wait before an answer begins. The rank favors consistent results across several models, while the table shows the published response-start time for practical expectations.
LLM Providers: Latency Leaderboard & Benchmarks
The model-adjusted rating compares consistent first-token latency across exact-model comparisons, so one unusually strong model cannot dominate the result. A model-adjusted rating requires at least three exact models measured on another provider. Providers with only their own models remain visible using raw published catalog speed.
- Models reviewed:
- 98
- Model-adjusted providers:
- 43
Scope
Rank by
| Position | Provider | Speed measurements | Data coverage |
|---|---|---|---|
| Rank | |||
| #1 | Speed measurements
| Data coverage
| |
| #2 | Speed measurements
| Data coverage
| |
| #3 | Speed measurements
| Data coverage
| |
| #4 | Speed measurements
| Data coverage
| |
| #5 | Speed measurements
| Data coverage
| |
| #6 | Speed measurements
| Data coverage
| |
| #7 | Speed measurements
| Data coverage
| |
| #8 | Speed measurements
| Data coverage
| |
| #9 | Speed measurements
| Data coverage
| |
| #10 | Speed measurements
| Data coverage
| |
| #11 | Speed measurements
| Data coverage
| |
| #12 | Speed measurements
| Data coverage
| |
| #13 | Speed measurements
| Data coverage
| |
| #14 | Speed measurements
| Data coverage
| |
| #15 | Speed measurements
| Data coverage
| |
| #16 | Speed measurements
| Data coverage
| |
| #17 | Speed measurements
| Data coverage
| |
| #18 | Speed measurements
| Data coverage
| |
| #19 | Speed measurements
| Data coverage
| |
| #20 | Speed measurements
| Data coverage
| |
| #21 | Speed measurements
| Data coverage
| |
| #22 | Speed measurements
| Data coverage
| |
| #23 | Speed measurements
| Data coverage
| |
| #24 | Speed measurements
| Data coverage
| |
| #25 | Speed measurements
| Data coverage
| |
| #26 | Speed measurements
| Data coverage
| |
| #27 | Speed measurements
| Data coverage
| |
| #28 | Speed measurements
| Data coverage
| |
| #29 | Speed measurements
| Data coverage
| |
| #30 | Speed measurements
| Data coverage
| |
| #31 | Speed measurements
| Data coverage
| |
| #32 | Speed measurements
| Data coverage
| |
| #33 | Speed measurements
| Data coverage
| |
| #34 | Speed measurements
| Data coverage
| |
| #35 | Speed measurements
| Data coverage
| |
| #36 | Speed measurements
| Data coverage
| |
| #37 | Speed measurements
| Data coverage
| |
| #38 | Speed measurements
| Data coverage
| |
| #39 | Speed measurements
| Data coverage
| |
| #40 | Speed measurements
| Data coverage
| |
| #41 | Speed measurements
| Data coverage
| |
| #42 | Speed measurements
| Data coverage
| |
| #43 | Speed measurements
| Data coverage
| |
| #44 | Speed measurements
| Data coverage
| |
| #45 | Speed measurements
| Data coverage
| |
| #46 | Speed measurements
| Data coverage
| |
| #47 | Speed measurements
| Data coverage
| |
| #48 | Speed measurements
| Data coverage
| |
| Without enough measurements · 44 providers | |||
| — | Ambient2 shared measured models; at least 3 required | Speed measurements
| Data coverage
|
| — | Decart2 shared measured models; at least 3 required | Speed measurements
| Data coverage
|
| — |  DeepSeek2 shared measured models; at least 3 required | Speed measurements
| Data coverage
|
| — | Nex AGI1 shared measured model; at least 3 required | Speed measurements
| Data coverage
|
| — | PoolsideNot enough shared-model measurements | Speed measurements
| Data coverage
|
| — | Reka AINot enough shared-model measurements | Speed measurements
| Data coverage
|
| — | RelaceNot enough shared-model measurements | Speed measurements
| Data coverage
|
| — | Xiaomi2 shared measured models; at least 3 required | Speed measurements
| Data coverage
|
| — | AI21Not enough shared-model measurements | Speed measurements
| Data coverage
|
| — | Arcee AINot enough shared-model measurements | Speed measurements
| Data coverage
|
| — |  InceptionNot enough shared-model measurements | Speed measurements
| Data coverage
|
| — | Ionstream1 shared measured model; at least 3 required | Speed measurements
| Data coverage
|
| — | OpenInference1 shared measured model; at least 3 required | Speed measurements
| Data coverage
|
| — | SakanaNot enough shared-model measurements | Speed measurements
| Data coverage
|
| — | StepFun1 shared measured model; at least 3 required | Speed measurements
| Data coverage
|
| — | UpstageNot enough shared-model measurements | Speed measurements
| Data coverage
|
| — | Amazon NovaNo models have complete speed data | Speed measurements
| Data coverage
|
| — | AvianNo models have complete speed data | Speed measurements
| Data coverage
|
| — | Black Forest LabsNo models have complete speed data | Speed measurements
| Data coverage
|
| — | CirrascaleNo models have complete speed data | Speed measurements
| Data coverage
|
| — | ClarifaiNo models have complete speed data | Speed measurements
| Data coverage
|
| — | CrucibleNo models have complete speed data | Speed measurements
| Data coverage
|
| — | CrusoeNo models have complete speed data | Speed measurements
| Data coverage
|
| — |  DarkbloomNo models have complete speed data | Speed measurements
| Data coverage
|
| — | DeepgramNo models have complete speed data | Speed measurements
| Data coverage
|
| — | FeatherlessNo models have complete speed data | Speed measurements
| Data coverage
|
| — | HeyGenNo models have complete speed data | Speed measurements
| Data coverage
|
| — | Inferact vLLMNo models have complete speed data | Speed measurements
| Data coverage
|
| — | InferenceNetNo models have complete speed data | Speed measurements
| Data coverage
|
| — | InfermaticNo models have complete speed data | Speed measurements
| Data coverage
|
| — | InflectionNo models have complete speed data | Speed measurements
| Data coverage
|
| — | LiquidNo models have complete speed data | Speed measurements
| Data coverage
|
| — | MetaNo models have complete speed data | Speed measurements
| Data coverage
|
| — | ModularNo models have complete speed data | Speed measurements
| Data coverage
|
| — | NCompassNo models have complete speed data | Speed measurements
| Data coverage
|
| — | NVIDIANo models have complete speed data | Speed measurements
| Data coverage
|
| — | PerceptronNo models have complete speed data | Speed measurements
| Data coverage
|
| — | QuiverNo models have complete speed data | Speed measurements
| Data coverage
|
| — |  RecraftNo models have complete speed data | Speed measurements
| Data coverage
|
| — | Sail ResearchNo models have complete speed data | Speed measurements
| Data coverage
|
| — | SourcefulNo models have complete speed data | Speed measurements
| Data coverage
|
| — | StealthNo models have complete speed data | Speed measurements
| Data coverage
|
| — | SwitchpointNo models have complete speed data | Speed measurements
| Data coverage
|
| — | TenstorrentNo models have complete speed data | Speed measurements
| Data coverage
|
Ranking data fetched .
Provider benchmark guide
How to use the LLM Providers: Latency Leaderboard & Benchmarks
43 providers currently have enough shared measurements for this benchmark. Weights & Biases is currently first, with a published catalog median of 0.35 seconds. Review response time, response start, output speed, and coverage together before choosing a provider.
How provider catalogs are represented
This overall view reviewed 98 candidate models selected across provider catalogs. Each observation is normalized against other providers running the exact same model.
Read the coverage before choosing
This page reviewed 98 candidate models and currently has shared comparison data for 43 providers. The Rating basis field distinguishes model-adjusted rows from providers positioned using only their own-model catalog measurements.