Best published cost/speed balance
Groq5 priced · 5 speed-measured models
$0.3 / 1M blended tokens · 1.98 s median response
Across 5 priced · 5 speed-measured models
Independent inference performance data
LLM Providers: Leaderboard & Inference Benchmarks
Compare same-model response time, first-token speed, output speed, price, reliability, available models, context size, and exact OpenRouter routes.
At a glance
Published catalog winners
These cards use each visible provider’s median published catalog price and response time; they are separate from the model-adjusted leaderboard rating. Prices use a 1,000-input/500-output-token mix normalized to 1 million total tokens.
Lowest published price
DekaLLM5 priced · 5 speed-measured models
$0.08 / 1M blended tokens
Compared across 5 priced models
Fastest published catalog
Groq5 priced · 5 speed-measured models
1.98 s median response
Compared across 5 measured models
Best published cost/speed balance
Groq5 priced · 5 speed-measured models
$0.3 / 1M blended tokens · 1.98 s median response
Across 5 priced · 5 speed-measured models
Lowest published price
DekaLLM5 priced · 5 speed-measured models
$0.08 / 1M blended tokens
Compared across 5 priced models
Fastest published catalog
Groq5 priced · 5 speed-measured models
1.98 s median response
Compared across 5 measured models
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
|
| — | 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
|
| — | PerceptronNot enough shared-model measurements | Speed measurements
| Data coverage
|
| — | RelaceNot enough shared-model measurements | 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
|
| — | 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 . See the popular-model weekly ranking.
Practical comparison guide
How to compare LLM inference providers
The model is only one part of an AI application’s performance. The inference provider determines where that model runs, how quickly a request begins, how fast tokens arrive, what the request costs, and which context limits and routing options are available. This leaderboard brings those operational differences together so teams can create a practical shortlist before testing providers with their own prompts and traffic patterns.
Separate latency from output speed
Response starts measures the wait for the first token, which matters for chat and interactive tools. Output speed measures generated tokens per second, which matters more for long answers. The 500-token response estimate combines both into one user-friendly duration.
Compare price with reliability
A low token price does not automatically make an endpoint the best production choice. Review input and output pricing alongside recent uptime, context size, model availability, and the exact routing tag your integration would use.
Understand what supports a rank
Providers need at least three complete speed measurements to appear by default. The ranking rewards consistent performance across several models and gives broader evidence more weight. Providers that offer only proprietary models remain in the primary list.