Provider reliability · tracked daily
Which provider is most reliable for each model?
Uptime and latency, per provider, per model - measured directly from OpenRouter's own endpoint data. The uptime figure is OpenRouter's own trailing-30-minute metric, captured once daily by our sync (not a live/real-time feed) - a daily snapshot of that window, not this instant. A model's benchmark score doesn't tell you whether you can actually depend on getting a response; this does.
Which host is most reliable overall?
Alibaba leads among hosts serving 3+ tracked models, at 100.0% median uptime across 5 models. Inverted from the table above: instead of "which host is best for this model," this ranks each host by measured reliability across every model it serves.
| # | Provider | Models served | Median uptime | Avg uptime | Peak throughput |
|---|---|---|---|---|---|
| 01 | Alibaba | 5 | 100.0% | 99.5% | 73 tok/s |
| 02 | Cloudflare | 4 | 100.0% | 100.0% | 77 tok/s |
| 03 | Cerebras | 2 | 100.0% | 100.0% | 777 tok/s |
| 04 | Mara | 2 | 100.0% | 100.0% | 246 tok/s |
| 05 | Poolside | 2 | 100.0% | 100.0% | 156 tok/s |
| 06 | Inceptron | 1 | 100.0% | 100.0% | 45 tok/s |
| 07 | Sail Research | 1 | 100.0% | 100.0% | 31 tok/s |
| 08 | xAI | 1 | 100.0% | 100.0% | 50 tok/s |
| 09 | Azure | 13 | 100.0% | 97.5% | 69 tok/s |
| 10 | DeepSeek | 2 | 100.0% | 100.0% | 80 tok/s |
| 11 | Groq | 2 | 100.0% | 100.0% | 367 tok/s |
| 12 | Moonshot AI | 1 | 100.0% | 100.0% | 27 tok/s |
| 13 | Friendli | 3 | 100.0% | 100.0% | 115 tok/s |
| 14 | ModelRun | 1 | 100.0% | 100.0% | 129 tok/s |
| 15 | Tencent | 1 | 100.0% | 100.0% | 46 tok/s |
| 16 | Anthropic | 7 | 100.0% | 97.9% | 69 tok/s |
| 17 | Io Net | 1 | 100.0% | 100.0% | 11 tok/s |
| 18 | Decart | 1 | 100.0% | 100.0% | 60 tok/s |
| 19 | Amazon Bedrock | 11 | 99.9% | 97.7% | 150 tok/s |
| 20 | Baidu | 5 | 99.9% | 99.7% | 80 tok/s |
| 21 | Novita | 15 | 99.9% | 98.3% | 78 tok/s |
| 22 | Ionstream | 2 | 99.8% | 99.8% | 25 tok/s |
| 23 | Ambient | 2 | 99.8% | 99.8% | 88 tok/s |
| 24 | AkashML | 2 | 99.8% | 99.8% | 57 tok/s |
| 25 | Z.AI | 1 | 99.8% | 99.8% | 49 tok/s |
| 26 | AtlasCloud | 8 | 99.7% | 99.6% | 49 tok/s |
| 27 | SiliconFlow | 7 | 99.5% | 98.2% | 65 tok/s |
| 28 | Venice | 9 | 99.5% | 94.3% | 83 tok/s |
| 29 | DeepInfra | 17 | 99.4% | 98.4% | 137 tok/s |
| 30 | Crusoe | 2 | 99.4% | 99.4% | 99 tok/s |
| 31 | BaseTen | 5 | 99.3% | 96.4% | 189 tok/s |
| 32 | 15 | 99.3% | 97.1% | 120 tok/s | |
| 33 | Modal | 1 | 99.3% | 99.3% | 60 tok/s |
| 34 | DigitalOcean | 8 | 99.3% | 98.8% | 106 tok/s |
| 35 | Parasail | 9 | 99.2% | 98.6% | 123 tok/s |
| 36 | GMICloud | 9 | 99.2% | 92.5% | 88 tok/s |
| 37 | Fireworks | 5 | 99.2% | 97.4% | 147 tok/s |
| 38 | Xiaomi | 2 | 99.2% | 99.2% | 47 tok/s |
| 39 | Google AI Studio | 5 | 99.0% | 98.8% | 139 tok/s |
| 40 | OpenAI | 6 | 98.9% | 98.7% | 154 tok/s |
| 41 | CoreWeave | 5 | 98.8% | 98.2% | 106 tok/s |
| 42 | Chutes | 2 | 98.7% | 98.7% | 25 tok/s |
| 43 | Wafer | 2 | 98.6% | 98.6% | 43 tok/s |
| 44 | Together | 7 | 98.6% | 89.9% | 128 tok/s |
| 45 | StreamLake | 5 | 98.5% | 97.9% | 35 tok/s |
| 46 | StepFun | 1 | 98.1% | 98.1% | 47 tok/s |
| 47 | Morph | 5 | 98.0% | 97.4% | 35 tok/s |
| 48 | Minimax | 2 | 97.7% | 97.7% | 59 tok/s |
| 49 | Nebius | 2 | 97.7% | 97.7% | 183 tok/s |
| 50 | SambaNova | 4 | 97.6% | 97.6% | 359 tok/s |
| 51 | Mancer 2 | 2 | 96.8% | 96.8% | 41 tok/s |
| 52 | OpenInference | 2 | 95.2% | 95.2% | 14 tok/s |
| 53 | NextBit | 1 | 94.3% | 94.3% | 40 tok/s |
| 54 | Phala | 5 | 89.5% | 67.2% | 88 tok/s |
| 55 | Claude Platform on AWS | 1 | - | - | - |