Models · Mistral · live prices
Mixtral 8x7B: GPUs needed and cost to self-host
46.7B parameters, 12.9B active per token (mixture of experts: every expert sits in memory, but only the active ones are read per token, so it decodes like a 12.9B model). In FP8 that is about 56 GB with room for the KV cache.
Memory, FP8
56 GB
weights + ~20%
Cheapest per 1M output tokens
$0.04
1× A100 · on-demand · 60% utilization
Hosted API, sample list price
$0.60
per 1M output tokens
Cost per 1M output tokens, FP8
Prices observed 11m ago. Cheapest on-demand offer per GPU, whole instances, 60% average utilization.
- A10080 GB 1× on Vast.ai · on-demand · $0.34/hr · ~3,793 tok/s $0.04per 1M out
- H200141 GB 1× on Vast.ai · on-demand · $1.32/hr · ~8,930 tok/s $0.07per 1M out
- 509032 GB 2× on Vast.ai · on-demand · $0.92/hr · ~4,001 tok/s $0.11per 1M out
- 409024 GB 4× on Vast.ai · on-demand · $1.24/hr · ~4,501 tok/s $0.13per 1M out
- RTX PRO 600096 GB 1× on Vast.ai · on-demand · $1.00/hr · ~3,334 tok/s $0.14per 1M out
- H10080 GB 1× on RunPod · community · $1.99/hr · ~6,233 tok/s $0.15per 1M out
- B300288 GB 1× on RunPod · community · $6.94/hr · ~14,884 tok/s $0.22per 1M out
The tick marks the sample hosted-API price. Estimates: tokens/s from memory bandwidth with batched serving, not a benchmark. Idle hours still bill, so lower utilization means pricier tokens.
GPUs per replica by precision
One node, split 1, 2, 4 or 8 ways; 90% of VRAM usable.
| GPU | BF16112 GB | FP856 GB | INT428 GB |
|---|---|---|---|
| H10080 GB | 2× | 1× | 1× |
| H200141 GB | 1× | 1× | 1× |
| B300288 GB | 1× | 1× | 1× |
| A10080 GB | 2× | 1× | 1× |
| RTX PRO 600096 GB | 2× | 1× | 1× |
| 509032 GB | 4× | 2× | 1× |
| 409024 GB | 8× | 4× | 2× |