Where to run open models
Some open-weight models are served by more than one provider at very different prices, because the weights are the same but the host is not. This page collects every family with more than one host and ranks them cheapest first; open a family for the full breakdown and the real monthly gap.
9 families, 29 hosted listings
| Family | Providers | Cheapest | Cheapest blended price | Spread |
|---|---|---|---|---|
| Llama 3.1 8B | 3 | DeepInfra | $0.024 | 7.5× spread |
| Llama 3.3 70B | 3 | DeepInfra | $0.144 | 7.2× spread |
| gpt-oss-120b | 3 | OpenRouter | $0.0636 | 3.8× spread |
| DeepSeek V4 Pro | 4 | OpenRouter | $0.817034 | 2.6× spread |
| gpt-oss-20b | 3 | OpenRouter | $0.05 | 2.4× spread |
| Llama 4 Maverick | 3 | DeepInfra | $0.32 | 1.2× spread |
| Kimi K3 | 4 | DeepInfra | $5.13 | 1.1× spread |
| GLM-5.3 | 3 | Fireworks AI | $2.00 | same everywhere |
| Qwen3.8 Max | 3 | Alibaba Qwen | $2.80 | same everywhere |