family member
Ministral 3 logo

Ministral 3 3B-INSTRUCT-2512

17 tokens a second on Linux · L4 23 GB · 12 CPUs · 53 GB RAM. About $13 per million tokens here, against OpenRouter's $0.10: the API is 132× cheaper. Run it here for privacy or to keep work on your own machines, not to save money.

1 build of Ministral 3 3B-INSTRUCT-2512, 3 signed runs on real machines: different packagers, precisions and engines, each a different file with its own numbers.

17tok/s, fastest machine
$13per M tokens · $0.10 on an API
64Kcontext proved by recall
35.1 GBmemory, smallest build
3providers measured it
Run it on your gram Install a gram with one command. It works through the catalog on your own machine and signs every result. 3 providers have measured Ministral 3 3B-INSTRUCT-2512 so far.Use it in your own room Ask Evo in your gramx and it runs on a machine that proved it, at the metered $13 per million tokens. On your own machines it costs you nothing.Compare every build Put it beside every other model in one table: speed, price, context and the API price, one row each.
best on ollama · 1 build

trained 256K · proved 64K context, by mistralai Q4_K_M

16.5 tok/s peak, by mistralai Q4_K_M

Speed against price, every machine

Up and left is faster and cheaper · the line joins machines no other beats on both · price on a log scale · dashed: OpenRouter

$0.1$1$10USD per million tokenstok/sLinux · L4 23 GB · 12 CPUs · 53 GB RAM · 4-bit by mistralai · 16.51 tok/s · $13.1616/MWindows · Intel(R) Iris(R) Xe Graphics 1 GB · 8 CPUs · 16 GB RAM · 4-bit by mistralai · 4.66 tok/s · $15.5973/M

Same build, different machines

Q4_K_M by mistralai on 3 machines: the fastest decodes 6.0× as fast as the slowest. One dot per machine, tokens per second.

  1. NVIDIA L4×117 tok/s
  2. Intel(R) Iris(R) Xe Graphics×15.1 tok/s
  3. AMD Radeon(TM) Vega 8 Mobile Graphics×12.8 tok/s

What it fits on

The smallest build needs about 35.1 GB; 1 of 18 kinds of machine on the network can hold it. Bars: each kind's memory (its card's, else RAM) · ticks: every build, at the size it held plus a fifth for the prompt.

  1. Linux, CPU only · 2 CPUs1 GB RAMnone
  2. NVIDIA GeForce RTX 3050 Laptop GPU4 GB on the cardnone
  3. NVIDIA GeForce GTX 16504 GB on the cardnone
  4. AMD Radeon(TM) Vega 8 Mobile Graphics7 GB RAMnone
  5. Windows, CPU only · 12 CPUs7 GB RAMnone
  6. Apple silicon8 GB RAMnone
  7. Apple M18 GB RAMnone
  8. Mac, CPU only · 8 CPUs8 GB RAMnone
  9. NVIDIA GeForce RTX 3080 Ti12 GB on the cardnone
  10. Tesla T415 GB on the cardnone
  11. Intel(R) UHD Graphics 73016 GB RAMnone
  12. Intel(R) Iris(R) Xe Graphics16 GB RAMnone
  13. Apple M416 GB RAMnone
  14. Apple M116 GB RAMnone
  15. NVIDIA L422 GB on the cardnone
  16. 2× Tesla T429 GB on the cardnone
  17. Linux, CPU only · 4 CPUs31 GB RAMnone
  18. NVIDIA RTX PRO 6000 Blackwell Server Edition95 GB on the cardevery build

How far back it remembers, machine by machine

Filled: the needle came back at that prompt size · outlined: asked and lost · blank: not asked · trained for 256K. 0 walks ended because the model forgot, 2 because the machine could not hold the prompt.

machine · build4K8K16K32K64Kwhy it stopped
Gram ·dae679NVIDIA L4 · Q4_K_M by mistralai64K too slow to answer here
Gram ·2e5c02Intel(R) Iris(R) Xe Graphics · Q4_K_M by mistralai4K this machine could not serve at 8K

Carbon per million tokens

Lowest build on each machine · 1 of 2 read off a power sensor, the rest modelled from the parts · fewer grams is better

  1. Gram ·dae6793.6 kg27.75 J/tok · sensor
  2. Gram ·2e5c026.8 kg34.40 J/tok · modelled

Carbon is location-based: signed energy × the yearly average of the grid where the machine signed its place (world grid average (Singapore not in the table) · India grid average; Ember Global Electricity Review 2025 (2024 generation), location-based). Facility overhead is not included. A machine on its own solar is not credited until it can prove it.

Measured by

The people whose machines signed these numbers

proved by at least one build: Tool calling · Sustained output · Vision · Counts in images · Multiple images at once · Spatial reasoning · Writes code · Follows instructions

Every build

Each its own file with its own page: the machines that ran it and their signatures.

buildenginewhose workmemorycontext provedbest tok/smachines
Q4_K_M by mistralaihf.co/mistralai/Ministral-3-3B-Instruct-2512-GGUF:Q4_K_Mollamavendor · mistralai35.1 GB64K16.53
derivatives — somebody else's change to these weights

Fine-tunes, merges and abliterations of this member, each its own model with its own page. None of their numbers are counted above.

Ministral-3-3B-Instruct-2512not measured