Every host for Llama 3.1 8B Instruct
| Host | Blended | Input | Output | Cached | Precision | Uptime |
|---|---|---|---|---|---|---|
| DeepInfra | $0.025 | $0.02 | $0.04 | — | fp8 | 99.9% |
| Novita | $0.028 | $0.02 | $0.05 | — | fp8 | 99.1% |
| Nscale | $0.03 | $0.03 | $0.03 | — | — | — |
| Groq | $0.058 | $0.05 | $0.08 | $0.025 | — | 100.0% |
| Ovhcloud | $0.10 | $0.10 | $0.10 | — | — | — |
| Cloudflare | $0.186 | $0.152 | $0.287 | — | fp8 | 98.3% |
| Perplexity | $0.20 | $0.20 | $0.20 | — | — | — |
| CoreWeave | $0.22 | $0.22 | $0.22 | $0.22 | bf16 | 99.9% |
| Weights & Biases | $0.22 | $0.22 | $0.22 | — | — | — |
| Inferenceapis | $0.275 | $0.20 | $0.50 | — | — | — |
Cheapest price over time
- Input
- Output
What your workload costs
- DeepInfra$10.00
- Novita$11.00
- Nscale$12.00
- Groq$23.00
- Ovhcloud$40.00
- Cloudflare$74.30
- Perplexity$80.00
- CoreWeave$88.00
- Weights & Biases$88.00
- Inferenceapis$110.00
Rent GPUs or pay per token?
- Renting the GPUs
- $16,878/mo
- Break-even volume
- 675B tokens/mo
- Needed around the clock
- 257K tok/s
8 × H100 cost $16,878 a month. The cheapest API (DeepInfra) charges $0.025 per 1M blended tokens, so the GPUs only pay off above 675B tokens a month: the box must serve 257K tokens a second, every second.
GPU price: cheapest on-demand offer per GPU-hour × 730 hours. No throughput is assumed: it depends on the model size, precision, batch size and serving software, so measure yours. Power, storage, egress and engineering time are not included.
LMArena rating over time
Rated about the same, by price
Quality data: LMArena (CC BY 4.0), Epoch AI (CC BY 4.0). Prices: OpenRouter, LiteLLM (MIT), provider pages. List prices in US dollars before discounts and taxes; other currencies converted at the European Central Bank reference rate. The figures on these pages are free to reuse under CC BY 4.0 with a link to this page.