FitLocally by GPUwerk

Mellum · 12.1B · MoE

Can you run Mellum2.1 12B-A2.5B Thinking locally?

Mellum2.1 12B-A2.5B Thinking is a mixture-of-experts model with 12.1B parameters, 2.5B active per token. At 4-bit or better it runs on 112 of the 120 hardware configurations listed here; the smallest is the RTX 5070 (12 GB). On a 128 GB DGX Spark it runs at Q4_K_M, at roughly 90 tokens per second (our est.).

Data as of 2 November 2026

Fastest machines for Mellum2.1 12B-A2.5B Thinking

Every machine that runs it, fastest first. Filter by type; your machine is marked. Compare machines running it side by side

  1. 1RTX 5090Graphics card · 32 GB~589 tok/s our est.Q4_K_M · 9.3 of 31 GB
  2. 22x RTX 5090Graphics card · 64 GB~589 tok/s our est.Q4_K_M · 9.3 of 62 GB
  3. 3RTX PRO 6000 BlackwellGraphics card · 96 GB~589 tok/s our est.Q4_K_M · 9.3 of 95 GB
  4. 42x RTX PRO 6000 BlackwellGraphics card · 192 GB~589 tok/s our est.Q4_K_M · 9.3 of 190 GB
  5. 5RTX 4090Graphics card · 24 GB~331 tok/s our est.Q4_K_M · 9.3 of 23 GB
  6. 6RTX 3090 TiGraphics card · 24 GB~331 tok/s our est.Q4_K_M · 9.3 of 23 GB
  7. 72x RTX 4090Graphics card · 48 GB~331 tok/s our est.Q4_K_M · 9.3 of 46 GB
  8. 82x RTX 3090 TiGraphics card · 48 GB~331 tok/s our est.Q4_K_M · 9.3 of 46 GB
  9. 9RTX 5080Graphics card · 16 GB~315 tok/s our est.Q4_K_M · 9.3 of 15.0 GB
  10. 10RX 7900 XTXGraphics card · 24 GB~315 tok/s our est.Q4_K_M · 9.3 of 23 GB
  11. 112x RX 7900 XTXGraphics card · 48 GB~315 tok/s our est.Q4_K_M · 9.3 of 46 GB
  12. 12RTX 3090Graphics card · 24 GB~308 tok/s our est.Q4_K_M · 9.3 of 23 GB
  13. 132x RTX 3090Graphics card · 48 GB~308 tok/s our est.Q4_K_M · 9.3 of 46 GB
  14. 14RTX 5070 TiGraphics card · 16 GB~294 tok/s our est.Q4_K_M · 9.3 of 15.0 GB
  15. 15M3 UltraMac · 96–512 GB · 5 variants~269 tok/s our est.Q4_K_M · 9.3 GB
  16. 16RX 7900 XTGraphics card · 20 GB~263 tok/s our est.Q4_K_M · 9.3 of 19.0 GB
  17. 17M2 UltraMac · 64–192 GB · 7 variants~263 tok/s our est.Q4_K_M · 9.3 GB
  18. 18M1 UltraMac · 64–128 GB · 4 variants~263 tok/s our est.Q4_K_M · 9.3 GB
  19. 19RTX 4080 SUPERGraphics card · 16 GB~242 tok/s our est.Q4_K_M · 9.3 of 15.0 GB
  20. 20Tesla P100Graphics card · 16 GB~241 tok/s our est.Q4_K_M · 9.3 of 15.0 GB
  21. 21RTX 4080Graphics card · 16 GB~236 tok/s our est.Q4_K_M · 9.3 of 15.0 GB
  22. 22RTX 5070Graphics card · 12 GB~221 tok/s our est.Q4_K_M · 9.3 of 11.0 GB
  23. 23RTX 4070 Ti SUPERGraphics card · 16 GB~221 tok/s our est.Q4_K_M · 9.3 of 15.0 GB
  24. 24RX 9070 XTGraphics card · 16 GB~210 tok/s our est.Q4_K_M · 9.3 of 15.0 GB
  25. 25M4 MaxMac · 48–128 GB · 3 variants~179 tok/s our est.Q4_K_M · 9.3 GB
  26. 26RTX 4070 TiGraphics card · 12 GB~166 tok/s our est.Q4_K_M · 9.3 of 11.0 GB
  27. 27RTX 4070 SUPERGraphics card · 12 GB~166 tok/s our est.Q4_K_M · 9.3 of 11.0 GB
  28. 28RTX 4070Graphics card · 12 GB~166 tok/s our est.Q4_K_M · 9.3 of 11.0 GB
  29. 29RTX 5060 TiGraphics card · 16 GB~147 tok/s our est.Q4_K_M · 9.3 of 15.0 GB
  30. 30M4 Max · 36 GB · 32cMac · 36 GB~135 tok/s our est.Q4_K_M · 9.3 of 24 GB
  31. 31M2 MaxMac · 32–96 GB · 5 variants~131 tok/s our est.Q4_K_M · 9.3 GB
  32. 32M1 MaxMac · 32–64 GB · 4 variants~131 tok/s our est.Q4_K_M · 9.3 GB
  33. 33M3 MaxMac · 48–128 GB · 3 variants~131 tok/s our est.Q4_K_M · 9.3 GB
  34. 34RTX 3060Graphics card · 12 GB~118 tok/s our est.Q4_K_M · 9.3 of 11.0 GB
  35. 35Tesla T4Graphics card · 16 GB~105 tok/s our est.Q4_K_M · 9.3 of 15.0 GB
  36. 36RX 9060 XTGraphics card · 16 GB~105 tok/s our est.Q4_K_M · 9.3 of 15.0 GB
  37. 37M3 MaxMac · 36–96 GB · 2 variants~99 tok/s our est.Q4_K_M · 9.3 GB
  38. 38RTX 4060 TiGraphics card · 16 GB~95 tok/s our est.Q4_K_M · 9.3 of 15.0 GB
  39. 39M4 ProMac · 24–64 GB · 6 variants~90 tok/s our est.Q4_K_M · 9.3 GB
  40. 40DGX Spark (GB10)AI mini PC · 128 GB~90 tok/s our est.Q4_K_M · 9.3 of 122 GB
  41. 41Ryzen AI Max+ 395AI mini PC · 64–128 GB · 2 variants~84 tok/s our est.Q4_K_M · 9.3 GB
  42. 42M2 ProMac · 16–32 GB · 2 variants~66 tok/s our est.Q4_K_M · 9.3 GB
  43. 43M1 ProMac · 16–32 GB · 2 variants~66 tok/s our est.Q4_K_M · 9.3 GB
  44. 44M5Mac · 16–32 GB · 6 variants~50 tok/s our est.Q4_K_M · 9.3 GB
  45. 45M3 ProMac · 18–36 GB · 2 variants~49 tok/s our est.Q4_K_M · 9.3 GB
  46. 46M4Mac · 16–32 GB · 3 variants~39 tok/s our est.Q4_K_M · 9.3 GB
  47. 47M3Mac · 16–24 GB · 2 variants~33 tok/s our est.Q4_K_M · 9.3 GB
  48. 48M2Mac · 16–24 GB · 2 variants~33 tok/s our est.Q4_K_M · 9.3 GB
  49. 49M1 · 16 GBMac · 16 GB~22 tok/s our est.Q4_K_M · 9.3 of 10.7 GB
  50. 50M6Mac · 16–32 GB · 5 variantsno estimateQ4_K_M · 9.3 GB
  51. 51M5 ProMac · 24–64 GB · 6 variantsno estimateQ4_K_M · 9.3 GB
  52. 52M5 MaxMac · 36–128 GB · 4 variantsno estimateQ4_K_M · 9.3 GB
  53. 53M5 UltraMac · 96–512 GB · 5 variantsno estimateQ4_K_M · 9.3 GB
Too big for 8 machines

RTX 5060, RTX 4060, RTX 3070, RTX 3060 Ti, RTX 2070, M3 · 8 GB, M2 · 8 GB, M1 · 8 GB. Mellum2.1 12B-A2.5B Thinking needs about 9.3 GB at its smallest build.

Speeds are single-user decode estimates, not measurements. "ModelFit est." comes from ModelFit's model fitted on published runs (CC BY 4.0); "our est." is memory bandwidth divided by bytes read per token, times 0.55. Memory is what the best build needs, against what the machine can give a model. Method

About the model

Parameters
12.1B
Active per token
2.5B
Type
Mixture of experts
Good for
Agentic coding, Small machines

Builds and how to run them

QuantizationLoadMin. RAMCommand
Q4_K_M8.1 GB16 GBGGUF or MLX build

Load, minimum RAM and Ollama commands: ModelFit, CC BY 4.0. Load is memory with the model loaded, before long contexts.

Questions

Can I run Mellum2.1 12B-A2.5B Thinking on a Mac?

Yes on a Mac with enough memory: the smallest Apple configuration listed here that runs it at 4-bit or better is the M2 Pro · 16 GB.

How much memory does Mellum2.1 12B-A2.5B Thinking need?

About 9.3 GB at the smallest build in our data, including a margin for the runtime and an 8K-token context. Longer contexts need more.

Does Mellum2.1 12B-A2.5B Thinking run on a DGX Spark?

Yes, at Q4_K_M. The Spark has 128 GB of unified memory, about 121.6 GB usable.