Inference providers: reliability and prices
The same model is often sold by several providers at different prices. Here is who is cheapest most often — and who stays up.
Checked Oct 21, 2026 · uptime from 1 day of samples
| Provider | Models | Cheapest for | Uptime, 30 days | Bad days |
|---|---|---|---|---|
| OpenAI | 93 | 35 | 97.5% | 1 |
| DeepInfra | 78 | 32 | 97.8% | 1 |
| Azure | 56 | 25 | 98.9% | 1 |
| 51 | 26 | 94.2% | 1 | |
| Novita | 48 | 10 | 96.7% | 1 |
| SiliconFlow | 41 | 7 | 97.5% | 1 |
| Parasail | 40 | 4 | 97.5% | 1 |
| Amazon Bedrock | 38 | 22 | 97.4% | 1 |
| Venice | 37 | 3 | 94.5% | 1 |
| Alibaba | 34 | 12 | 99.4% | 0 |
| Anthropic | 29 | 14 | 100% | 0 |
| GMICloud | 27 | 6 | 94.9% | 1 |
| Google AI Studio | 26 | 18 | 98.5% | 1 |
| Mistral | 26 | 27 | 99.6% | 0 |
| AtlasCloud | 25 | 1 | 98.9% | 1 |
| CoreWeave | 22 | 1 | 98.2% | 1 |
| Phala | 20 | — | 93.5% | 1 |
| Cloudflare | 19 | 1 | 99.7% | 0 |
| Together | 19 | 2 | 95.8% | 1 |
| StreamLake | 18 | 4 | 97.1% | 1 |
| DigitalOcean | 16 | 1 | 98.5% | 1 |
| Z.AI | 14 | 2 | 99.5% | 0 |
| Relace | 13 | 3 | 99.9% | 0 |
| Nebius | 13 | 2 | 96.2% | 1 |
| Claude Platform on AWS | 12 | 12 | 100% | 0 |
| BaseTen | 12 | 1 | 98% | 1 |
| Baidu | 11 | — | 99.5% | 0 |
| Wafer | 10 | 1 | 99.2% | 0 |
| DekaLLM | 10 | 1 | 98.9% | 1 |
| Mancer 2 | 9 | — | 99.4% | 0 |
| Reka | 9 | — | 98.7% | 1 |
| Crusoe | 9 | 1 | 98.1% | 1 |
| Darkbloom | 9 | 6 | 98.7% | 1 |
| Minimax | 9 | 3 | 99.6% | 0 |
| Sail Research | 8 | 6 | 100% | 0 |
| InferenceNet | 8 | 6 | 99% | 0 |
| Morph | 8 | — | 98.9% | 1 |
| xAI | 8 | 14 | 93.6% | 1 |
| Cohere | 7 | — | 99.7% | 0 |
| Friendli | 7 | — | 99.2% | 0 |
| Fireworks | 7 | 1 | 96.9% | 1 |
| AkashML | 7 | — | 99.8% | 0 |
| AionLabs | 6 | — | 99.9% | 0 |
| Seed | 6 | — | 100% | 0 |
| SambaNova | 6 | — | 98.9% | 1 |
| Inceptron | 6 | — | 98.1% | 1 |
| Makora | 6 | — | 87% | 1 |
| Modal | 6 | 1 | 98.3% | 1 |
| PrimeIntellect | 6 | — | 99% | 0 |
| NextBit | 6 | 1 | 93.4% | 1 |
| Io Net | 6 | 1 | 99.6% | 0 |
| Chutes | 6 | 3 | 94.2% | 1 |
| Groq | 6 | — | 99.6% | 0 |
| Meta | 6 | — | 100% | 0 |
| Nvidia | 6 | — | 66.1% | 1 |
| Decart | 5 | 2 | 99.8% | 0 |
| Perplexity | 5 | — | 100% | 0 |
| Tencent | 5 | 2 | 100% | 0 |
| Mara | 4 | — | 78.9% | 1 |
| ModelRun | 4 | — | 99.1% | 0 |
| Poolside | 4 | — | 100% | 0 |
| Sakana AI | 4 | — | 100% | 0 |
| Ionstream | 3 | — | 99.3% | 0 |
| OpenInference | 3 | 1 | 98.2% | 1 |
| Inception | 3 | — | 99.7% | 0 |
| Moonshot AI | 3 | — | 100% | 0 |
| Near AI | 3 | 1 | 99.7% | 0 |
| Upstage | 3 | 4 | 100% | 0 |
| Xiaomi | 3 | — | 99.9% | 0 |
| DeepSeek | 2 | — | 100% | 0 |
| Liquid | 2 | — | 88.2% | 1 |
| Nex AGI | 2 | — | 99.3% | 0 |
| Cerebras | 2 | — | 100% | 0 |
| Perceptron | 2 | — | 99.9% | 0 |
| Tenstorrent | 2 | — | 99.6% | 0 |
| ScaleDown | 2 | — | 99.9% | 0 |
| StepFun | 2 | 2 | 100% | 0 |
| Thinking Machines | 2 | — | 100% | 0 |
| Unbiased | 2 | — | 97.3% | 1 |
| Arcee AI | 1 | — | 100% | 0 |
| Featherless | 1 | 1 | 76.5% | 1 |
Most reliable providers
Providers with at least three models, by average uptime: Anthropic (100%), Claude Platform on AWS (100%), Sail Research (100%), Seed (100%), Meta (100%).
“Cheapest for” counts models with two or more paid providers where this provider has the lowest blended price. A bad day is one with average uptime under 99%. See the methodology.