Ministral 3 8B 2512.
Ministral 3 8B is an Apache 2.0 model with text and image input and a 256K context, built for edge and local use, fitting in 12GB of VRAM in FP8.
1× Nvidia GeForce RTX 3050 · $0.022/hr.
Most-aggressive quantisation we have a working recommendation for. Lower precision = less VRAM = cheaper hardware, at a small accuracy cost.
Cheapest hosted endpoints.
Variants in the Ministral family.
Ministral 8B is an 8B parameter model featuring a unique interleaved sliding-...
Ministral 3 14B Instruct 2512 is Mistral's Apache 2.0 image-and-text model, 1...
Ministral 3 3B is the smallest Ministral 3 model, Apache 2.0, with text and i...
Frequently asked.
How do I run Ministral 3 8B 2512?
Where can I access Ministral 3 8B 2512?
How much does it cost to run Ministral 3 8B 2512?
Is Ministral 3 8B 2512 open-source or proprietary?
Cheapest hardware per quantisation.
Each row is one quantisation tier (the same weights compressed differently). Lower precision → lower VRAM → cheaper hardware, at the cost of small accuracy loss. $/hr is the cheapest rate on offer now. A multi-card setup uses a real instance of that size when one is rented; otherwise it is the single-card rate times the count.
| Quantisation | Cheapest GPU config | Total VRAM | Live $/hr | tokens/sec | |
|---|---|---|---|---|---|
|
FP16
FP16 — half precision (default)
|
24 GB |
$0.15/hr
Nosana · on-demand
|
36.7 | Compare → | |
|
FP8
FP8 — 8-bit float (Hopper / Blackwell)
|
16 GB |
$0.019/hr
Clore.ai · on-demand
|
25.1 | Compare → | |
|
INT4
INT4 — 4-bit integer (~4× VRAM saving)
|
8 GB |
$0.022/hr
Clore.ai · on-demand
|
35.2 | Compare → |
What it costs per month across providers.
Estimate your monthly bill for Ministral 3 8B 2512 across every host that publishes per-token pricing. Slide your token volumes; the chart + table re-rank cheapest-first.
Cheapest provider on the left.
Total monthly cost — input + output tokens combined.
Bill breakdown.
Rent the GPU instead of paying per token.
For an open-weights model like Ministral 3 8B 2512, you can rent a GPU and serve inference yourself. The math: cheapest GPU rental × 730 hours/month. Power is included in a rental's hourly price.
Assumes the GPU runs 24/7 at ~85% utilisation. If your traffic is bursty, you'll pay less for the API and probably more for the GPU (idle hours still cost rental). The breakeven analysis lives on the Self-host vs API breakeven tool.
About Ministral 3 8B 2512.
Ministral 3 8B Instruct 2512 is the middle size of Mistral's Ministral 3 family, released under Apache 2.0. It pairs an 8.4B language model with a 0.4B vision encoder, so it reads images as well as text, with a 256K context window. It handles native function calling and JSON output across languages including English, French, Chinese and Arabic. Mistral aims it at local assistants, image and document description, and translation. In FP8 it fits in 12GB of VRAM. Images work best near a 1:1 aspect ratio.