API providers
Who actually serves the models, what they charge for them, and how fast they answer. Prices and endpoint telemetry are ingested from OpenRouter — we do not run our own provider benchmarks.
84 hosts serving tracked models
All 84 of 84 providers
| Provider | Models served | Mean output t/s | Mean TTFT | Lowest blended $/M |
|---|---|---|---|---|
OpenAI
· first party |
113 | 78 | 5.4s | $0.069 |
DeepInfra
|
92 | 48 | 1.5s | $0.022 |
Novita
|
88 | 56 | 1.8s | $0.000 |
Alibaba
· first party |
61 | 71 | 1.0s | $0.055 |
Google
· first party |
61 | 91 | 3.3s | $0.087 |
Azure
· first party |
59 | 71 | 3.7s | $0.14 |
| Pa Parasail | 46 | 63 | 0.8s | $0.030 |
| Si SiliconFlow | 43 | 44 | 1.6s | $0.075 |
Together
|
43 | 71 | 0.7s | $0.075 |
| AB Amazon Bedrock · first party | 39 | 80 | 1.6s | $0.061 |
| Ve Venice | 39 | 43 | 1.8s | $0.11 |
| At AtlasCloud | 36 | 56 | 2.0s | $0.000 |
Anthropic
· first party |
31 | 89 | 2.1s | $1.00 |
| Ph Phala | 30 | 58 | 1.8s | $0.068 |
| Co CoreWeave | 29 | 92 | 0.4s | $0.055 |
| Gm GMICloud | 29 | 50 | 2.7s | $0.000 |
| St StreamLake | 29 | 58 | 1.5s | $0.052 |
Google AI Studio
|
26 | 87 | 5.1s | $0.000 |
Mistral
· first party |
26 | 98 | 1.0s | $0.075 |
Fireworks
|
25 | 84 | 2.0s | $0.12 |
| Di DigitalOcean | 23 | 23 | 1.0s | $0.12 |
| Cl Cloudflare | 19 | 48 | 1.4s | $0.041 |
| Ne Nebius | 19 | 59 | 1.9s | $0.10 |
| Ne NextBit | 18 | 53 | 1.7s | $0.060 |
| IN Io Net | 16 | 45 | 1.6s | $0.047 |
| Re Reka · first party | 15 | 61 | 1.2s | $0.095 |
BaseTen
|
14 | 116 | 0.6s | $0.16 |
| Mo Morph | 14 | 327 | 2.4s | $0.10 |
Z.AI
· first party |
14 | 41 | 4.6s | $0.24 |
Baidu
· first party |
13 | 71 | 1.0s | $0.11 |
| Pr PrimeIntellect | 13 | 73 | 1.2s | $0.13 |
| Da Darkbloom | 12 | 32 | 2.4s | $0.000 |
| Re Relace | 12 | 79 | 0.8s | $0.15 |
| Ak AkashML | 11 | 59 | 1.0s | $0.040 |
| Ch Chutes | 11 | 23 | 2.9s | $0.18 |
| Cr Crusoe | 11 | 77 | 0.6s | $0.087 |
| Wa Wafer | 11 | 63 | 1.9s | $0.12 |
| CP Claude Platform on AWS | 11 | 52 | 2.4s | $4.00 |
| De DekaLLM | 11 | 55 | 0.9s | $0.021 |
Groq
|
9 | 278 | 0.5s | $0.058 |
| Io Ionstream | 9 | 44 | 1.5s | $0.17 |
| M2 Mancer 2 | 9 | 41 | 1.9s | $0.096 |
| SR Sail Research | 9 | 34 | 1.7s | $0.089 |
| In InferenceNet | 9 | 68 | 0.9s | $0.060 |
| De Decart | 8 | 88 | 1.6s | $0.000 |
| Fr Friendli | 8 | 96 | 1.0s | $0.20 |
Minimax
· first party |
8 | 58 | 1.0s | $0.42 |
Nvidia
· first party |
8 | 51 | 1.2s | $0.000 |
xAI
· first party |
8 | 125 | 2.1s | $1.25 |
| Ma Makora | 8 | 100 | 0.8s | $0.12 |
Cohere
· first party |
7 | 61 | 0.4s | $0.000 |
| In Inceptron | 7 | 48 | 1.1s | $0.20 |
| Mo ModelRun | 7 | 115 | 0.5s | $0.000 |
SambaNova
|
7 | 89 | 2.2s | $0.34 |
| Ai AionLabs | 6 | 49 | 1.8s | $0.88 |
DeepSeek
· first party |
6 | 82 | 1.0s | $0.26 |
| Ma Mara | 6 | 121 | 1.6s | $0.30 |
| NA Nex AGI | 6 | 74 | 1.8s | $0.000 |
| Op OpenInference | 6 | 32 | 1.9s | $0.000 |
| Po Poolside | 6 | 68 | 1.2s | $0.000 |
| Se Seed · first party | 6 | 66 | 1.1s | $0.13 |
Meta
· first party |
5 | 103 | 3.5s | $0.12 |
Perplexity
· first party |
5 | 56 | 21.1s | $1.00 |
Tencent
· first party |
5 | 76 | 1.9s | $0.077 |
| Xi Xiaomi · first party | 5 | 52 | 2.9s | $0.17 |
| Mo Modal | 5 | 106 | 1.0s | $0.24 |
| Am Ambient | 4 | 25 | 2.6s | $0.10 |
Cerebras
|
4 | 518 | 0.3s | $0.45 |
Moonshot AI
· first party |
4 | 46 | 2.7s | $1.20 |
| SA Sakana AI | 4 | 75 | 7.5s | $1.71 |
| In Inception | 3 | 277 | 1.4s | $0.068 |
Upstage
· first party |
3 | 74 | 1.2s | $0.087 |
| St Stealth | 3 | 37 | 5.7s | $0.000 |
| In Inflection | 2 | — | — | $4.38 |
| Pe Perceptron | 2 | 48 | 1.0s | $0.49 |
| TM Thinking Machines · first party | 2 | 60 | 1.5s | $0.000 |
| Un Unbiased | 2 | 74 | 3.1s | $1.40 |
| Te Tenstorrent | 2 | 60 | 0.9s | $0.20 |
| Ai AI21 · first party | 1 | — | — | $3.50 |
| AA Arcee AI | 1 | 188 | 0.5s | $0.39 |
| St StepFun | 1 | 102 | 3.6s | $0.44 |
| Li Liquid | 1 | 142 | 0.5s | $0.000 |
| NA Near AI | 1 | 17 | 3.3s | $0.17 |
| Kr Krea | 1 | 99 | 1.9s | $0.39 |
How to read this. A host appears here once we have seen it serve at least one tracked model.
Mean output t/s and mean TTFT average that host's measurements across every model it serves, so a host
that only serves small models will look faster than one carrying frontier models — compare hosts on a single model
page for a like-for-like reading. Lowest blended $/M is the cheapest blended rate that host publishes for any
model, at a 3:1 input:output blend. Figures are ingested from OpenRouter endpoint listings and telemetry, with
metadata enriched from models.dev. See methodology and
attribution.