Pricing

The per-million-token cost of the LLM service is computed according to each model's throughput with the following formula:

\[ \mathrm{Output\ pricing\ [€/Mtok]} = \mathrm{H100\ GPU\ rate\ [€/hour]} \times \frac{N_{\mathrm{GPU}}}{\mathrm{Output\ throughput\ [Mtok/s]} \times 3600 \mathrm{s/hour} \times 0.5} \]

\[ \mathrm{Output\ pricing\ [€/Mtok]} = \mathrm{H100\ GPU\ rate\ [€/hour]} \times \mathrm{Conversion\ factor\ [hour/Mtok]} \]

Pricing of input tokens is 20% of output. The throughput is computed as an average over multiple standard benchmarks. The table below shows the values for the throughput and conversion factor for the different models available (and coming soon). The example pricing is computed assuming a standard rate of 4 euros per H100 hour and rounding to the cent and it does not account for IVA.

Model	Avg Output Throughput [tok/s]	Number of GPUs	Conversion factor [hour/Mtok]	Example output pricing [€/Mtok]
✅ Qwen/Qwen3.5-122B-A10B-FP8	1778.705	2	0.624674193	2.50
✅ Qwen/Qwen3.6-35B-A3B-FP8	3171.73	1	0.175158527	0.70
✅ Qwen/Qwen3.8-27B-FP8	3116.31	1	0.178273521	0.71
✅ google/gemma-4-26B-A4B-it	5830.25	1	0.095288462	0.38
✅ openai/gpt-oss-120b	5995.415	2	0.185326806	0.74
🔜 DeepSeek/DeepSeek-V4-Flash-0731	2948.16	4	0.753765814	3.02