Cheapest place to run Llama 3.1 8B
Three providers serve Llama 3.1 8B. The dearest costs 7.5× what the cheapest does for the same tokens.
Every host, ranked cheapest first
| Provider | Input / 1M | Output / 1M | Blended (4:1) | Context window |
|---|---|---|---|---|
| DeepInfraCheapest | $0.02 | $0.04 | $0.024 | 128K |
| OpenRouter | $0.05 | $0.08 | $0.056 | 131K |
| Together AI | $0.18 | $0.18 | $0.18 | 128K |
Running 10M input / 2.5M output tokens a month costs $0.30 at DeepInfra and $2.25 at Together AI — a difference of $1.95 a month.
What this ranking does not tell you
These are list prices for the same open weights served by different providers, nothing more. A cheaper host is not automatically the better one: providers running the same model can still differ in context window, quantisation, throughput, rate limits and region availability, and none of those show up in a price column. Check the provider's own page for the details that matter for your workload before switching.