LocalFit
Submit a benchmark

Apple M4 Pro (20-core GPU), 48 GB: what runs best?

RTX 4090RTX 5090RTX 3090RTX 4060 Ti (16 GB)RTX 5070 TiRadeon RX 7900 XTXM4 Max 128 GBM3 Ultra 512 GBDGX Spark (128 GB)All devices

Apple · Apple Silicon Mac

Apple M4 Pro (20-core GPU), 48 GB

Memory48 GB unifiedBandwidth273 GB/s vendor specPlanned for models75% of unified memory (editable)

Unified memory

Context
KV cache
Mode
Advanced: reserves, bandwidth, workload, models

Defaults are editorial planning choices, not measurements: 1 GiB runtime + 0.5 GiB margin per device, 75% of unified memory, 8 GiB RAM for the OS, a vendor “24 GB” treated as 24 GiB (enter your driver’s available VRAM to override), 5 t/s minimum speed (research editorial default). Memory ledger

Apple M4 Pro (20-core GPU), 48 GB · 75% (36.0 GiB) planned for models · 273 GB/s8K contextF16 KVany backend≥ 5 t/s

Best plans Apple M4 Pro (20-core GPU), 48 GB

Best quality at ≥ 5 t/s

Qwen3.8-27B

33.7 AA index Xhigh

MLX 8-bitMLX-LMUnified memory

~7.8–8.0 t/s Estimated

Faster tradeoff

K2-Horizon-MoVA-36B-A4B

25.3 AA index single mode

Q5_K_Mllama.cpp MetalUnified memory

~14–40 t/s Estimated extrapolated

Next point on the quality/speed front; then Qwen3.6-35B-A3B.

Possibly better · speed not established

No higher-scoring model has a feasible plan here.

WeightsKV cacheReserveFreeFits* = a state term unsizedMeasuredlinks to its sourceEstimatedcalibrated or extrapolated range, ~Ceilingupper bound, ≤Unknownnever zero

Meets your 5 t/s floor 34

Best plan per model that reaches your floor by a matching measurement, a calibrated range, or an extrapolated estimate (labelled). Sorted by the selected quality metric, then speed.

  1. 1

    Qwen3.8-27B

    Qwen/Qwen3.8-27B

    2026-08-1427B dense

    AA index33.7Xhigh
    Plan & memory
    MLX 8-bitMLX-LMUnified memoryFits*
    Unified29.5 / 36.0 GiB
    Generate~7.8–8.0 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~31 s–35 sEstimated
  2. 2

    K2-Horizon-MoVA-36B-A4B

    IFM/K2-Horizon-MoVA-36B-A4B

    2026-09-0336B MoE · 4B active

    AA index25.3single mode
    Plan & memory
    Q5_K_Mllama.cpp MetalUnified memoryFits
    Unified27.6 / 36.0 GiB
    Generate~14–40 t/sEstimatedextrapolated≈ meets 5 t/s
    First token · 4K prompt—Unknown
  3. 3

    K2-Horizon-7B

    IFM/K2-Horizon-7B

    2026-09-039B tensors · dense

    AA index20.6single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits
    Unified11.5 / 36.0 GiB
    Generate~19–23 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~12 s–12 sEstimated
  4. 4

    Qwen3.6-35B-A3B

    Qwen/Qwen3.6-35B-A3B

    2026-04-1635B MoE · 3B active

    AA index18.2Reasoning
    Plan & memory
    Q5_K_Mllama.cpp MetalUnified memoryFits*
    Unified25.8 / 36.0 GiB
    Generate~22–65 t/sEstimatedextrapolated≈ meets 5 t/s
    First token · 4K prompt—Unknown
  5. 5

    Muse-Glimmer-30B

    meta-models/Muse-Glimmer-30B

    2026-08-1030B dense

    AA index17.5High
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits
    Unified29.3 / 36.0 GiB
    Generate~6.5–7.8 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~39 s–41 sEstimated
  6. 6

    K2-Horizon-3.7B

    IFM/K2-Horizon-3.7B

    2026-09-035.1B tensors · dense

    AA index15.6single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits
    Unified7.6 / 36.0 GiB
    Generate~32–39 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~6.6 s–7.0 sEstimated
  7. 7

    EXAONE-4.5-33B

    LGAI-EXAONE/EXAONE-4.5-33B

    2026-04-0933B dense

    AA index13.2 est.Reasoning
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits
    Unified35.5 / 36.0 GiB
    Generate~5.4–6.4 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~43 s–46 sEstimated
  8. 8

    Qwen3.5-4B

    Qwen/Qwen3.5-4B

    2026-03-024B dense

    AA index13.1 est.Reasoning
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits*
    Unified6.1 / 36.0 GiB
    Generate~41–49 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~5.2 s–5.5 sEstimated
  9. 9

    NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16

    nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16

    2026-08-1130B MoE · 3B active

    AA index12.9single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits*
    Unified32.8 / 36.0 GiB
    Generate~15–44 t/sEstimatedextrapolated≈ meets 5 t/s
    First token · 4K prompt—Unknown
  10. 10

    granite-4.2-30b

    ibm-granite/granite-4.2-30b

    2026-08-2530B dense

    AA index12.7single mode
    Plan & memory
    MLX 8-bitMLX-LMUnified memoryFits
    Unified32.5 / 36.0 GiB
    Generate~7.2–7.4 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~34 s–39 sEstimated
  11. 11

    MiniCPM5-2B

    openbmb/MiniCPM5-2B

    2026-09-072.5B dense

    AA index12.5single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryFits
    Unified6.5 / 36.0 GiB
    Generate~37–44 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~3.3 s–3.5 sEstimated
  12. 12

    Nemotron-Cascade-2-30B-A3B

    nvidia/Nemotron-Cascade-2-30B-A3B

    2026-03-1930B MoE · 3B active

    AA index11.7 est.single mode
    Plan & memory
    MLX 8-bitMLX-LMUnified memoryFits*
    Unified32.8 / 36.0 GiB
    Generate~23–46 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~4.0 s–5.1 sEstimated
  13. 13

    Qwen3.5-9B

    Qwen/Qwen3.5-9B

    2026-03-029B dense

    AA index11.2Reasoning
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits*
    Unified10.9 / 36.0 GiB
    Generate~20–23 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~12 s–12 sEstimated
  14. 14

    granite-4.2-8b

    ibm-granite/granite-4.2-8b

    2026-08-258B dense

    AA index11.1single mode
    Plan & memory
    MLX 8-bitMLX-LMUnified memoryFits
    Unified11.4 / 36.0 GiB
    Generate~23–24 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~9.1 s–10 sEstimated
  15. 15

    G9v3-3B

    ai9stars/G9v3-3B

    2026-07 (day unknown)3B dense

    AA index10.8 est.single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryFits
    Unified7.5 / 36.0 GiB
    Generate~31–37 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~3.9 s–4.1 sEstimated
  16. 16

    granite-4.2-3b

    ibm-granite/granite-4.2-3b

    2026-08-253B dense

    AA index9.1single mode
    Plan & memory
    MLX 8-bitMLX-LMUnified memoryFits
    Unified5.7 / 36.0 GiB
    Generate~55–57 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~3.4 s–3.9 sEstimated
  17. 17

    Tri-21B-Think

    trillionlabs/Tri-21B-Think

    2026-02 (day unknown)21B dense

    AA index9.0 est.single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits
    Unified23.6 / 36.0 GiB
    Generate~8.5–10 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~27 s–29 sEstimated
  18. 18

    LFM2.5-2.6B

    LiquidAI/LFM2.5-2.6B

    2026-08-042.7B dense

    AA index8.4 est.single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryFits*
    Unified6.7 / 36.0 GiB
    Generate~35–42 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~3.5 s–3.7 sEstimated
  19. 19

    LFM2.5-8B-A1B

    LiquidAI/LFM2.5-8B-A1B

    2026-05-288.3B MoE · 1.5B active

    AA index7.2 est.single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryFits*
    Unified17.4 / 36.0 GiB
    Generate~17–48 t/sEstimatedextrapolated≈ meets 5 t/s
    First token · 4K prompt—Unknown
  20. 20

    Olmo-3.1-32B-ThinkSpecialist

    allenai/Olmo-3.1-32B-Think

    2025-12-1232B tensors · dense

    AA index7.1 est.single mode
    Plan & memory
    MLX 8-bitMLX-LMUnified memoryFits
    Unified34.6 / 36.0 GiB
    Generate~6.6–6.8 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~37 s–42 sEstimated
  21. 21

    Qwen3.5-2B

    Qwen/Qwen3.5-2B

    2026-03-022B dense

    AA index6.9 est.Reasoning
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits*
    Unified3.5 / 36.0 GiB
    Generate~91–109 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~2.6 s–2.8 sEstimated
  22. 22

    Olmo-3.1-32B-Instruct

    allenai/Olmo-3.1-32B-Instruct

    2025-12-1232B tensors · dense

    AA index6.5 est.single mode
    Plan & memory
    MLX 8-bitMLX-LMUnified memoryFits
    Unified34.6 / 36.0 GiB
    Generate~6.6–6.8 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~37 s–42 sEstimated
  23. 23

    Qwen3.5-0.8B

    Qwen/Qwen3.5-0.8B

    2026-03-020.8B dense

    AA index6.1 est.Reasoning
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits*
    Unified2.4 / 36.0 GiB
    Generate~219–262 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~1.0 s–1.1 sEstimated
  24. 24

    granite-4.0-h-small

    ibm-granite/granite-4.0-h-small

    2025-10-0232B tensors · MoE

    AA index6.0 est.single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits*
    Unified33.5 / 36.0 GiB
    Generate~5.5–16 t/sEstimatedextrapolated≈ meets 5 t/s
    First token · 4K prompt—Unknown
  25. 25

    Phi-4Older

    microsoft/phi-4

    2024-12-1215B tensors · dense

    AA index5.9 est.single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryFits
    Unified30.4 / 36.0 GiB
    Generate~6.4–7.7 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~19 s–20 sEstimated
  26. 26

    Olmo-3-7B-ThinkSpecialist

    allenai/Olmo-3-7B-Think

    2025-11-207.3B tensors · dense

    AA index5.6 est.single mode
    Plan & memory
    MLX 8-bitMLX-LMUnified memoryFits
    Unified11.2 / 36.0 GiB
    Generate~23–24 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~8.3 s–9.4 sEstimated
  27. 27

    Olmo-3-7B-Instruct

    allenai/Olmo-3-7B-Instruct

    2025-11-207.3B tensors · dense

    AA index5.2 est.single mode
    Plan & memory
    MLX 8-bitMLX-LMUnified memoryFits
    Unified11.2 / 36.0 GiB
    Generate~23–24 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~8.3 s–9.4 sEstimated
  28. 28

    LFM2.5-1.2B-ThinkingSpecialist

    LiquidAI/LFM2.5-1.2B-Thinking

    2026-01-201.2B tensors · dense

    AA index5.2 est.single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryFits*
    Unified3.8 / 36.0 GiB
    Generate~81–97 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~1.5 s–1.6 sEstimated
  29. 29

    LFM2.5-1.2B-Instruct

    LiquidAI/LFM2.5-1.2B-Instruct

    2026-01-051.2B tensors · dense

    AA index5.2 est.single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryFits*
    Unified3.8 / 36.0 GiB
    Generate~81–97 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~1.5 s–1.6 sEstimated
  30. 30

    Molmo2-8BSpecialist

    allenai/Molmo2-8B

    2025-12-118.7B tensors · dense

    AA index5.0 est.single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryFits
    Unified17.9 / 36.0 GiB
    Generate~11–14 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~11 s–12 sEstimated
  31. 31

    LFM2.5-VL-1.6BSpecialist

    LiquidAI/LFM2.5-VL-1.6B

    2026-01-051.6B tensors · dense

    AA index4.8 est.single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryFits*
    Unified3.8 / 36.0 GiB
    Generate~81–97 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~2.1 s–2.2 sEstimated
  32. 32

    Apertus-8B-Instruct-2509

    swiss-ai/Apertus-8B-Instruct-2509

    2025-09-028.1B tensors · dense

    AA index4.8 est.single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryFits
    Unified17.5 / 36.0 GiB
    Generate~12–14 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~11 s–11 sEstimated
  33. 33

    NVIDIA-Nemotron-3-Nano-4B-BF16

    nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16

    2026-03-164B dense

    AA index4.8single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits*
    Unified5.6 / 36.0 GiB
    Generate~45–54 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~5.2 s–5.5 sEstimated
  34. 34

    K2-Horizon-0.9B

    IFM/K2-Horizon-0.9B

    2026-09-031.1B tensors · dense

    AA index3.0single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits
    Unified3.0 / 36.0 GiB
    Generate~140–167 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~1.4 s–1.5 sEstimated
+ 57 models without a AA index score

Memory not fully resolved 9

The weights fit, but the KV cache for this attention type (or the chosen cache layout) is not established, so a fit cannot be guaranteed.

  1. 1

    gemma-4-26B-A4B-it

    google/gemma-4-26B-A4B-it

    2026-04-0225B MoE · 3.8B active

    AA index16.7 est.Reasoning
    Plan & memory
    MLX 8-bitMLX-LMUnified memoryKV unknown
    Unified27.5+ / 36.0 GiB
    Generate~18–37 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~5.1 s–6.4 sEstimated
  2. 2

    gemma-4-31B-it

    google/gemma-4-31B-it

    2026-04-0231B dense

    AA index14.7Reasoning
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryKV unknown
    Unified31.9+ / 36.0 GiB
    Generate~5.9–7.1 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~40 s–42 sEstimated
  3. 3

    gemma-4-12B-it

    google/gemma-4-12B-it

    2026-06-0312B dense

    AA index14.2 est.Reasoning
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryKV unknown
    Unified13.3+ / 36.0 GiB
    Generate~15–18 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~16 s–17 sEstimated
  4. 4

    Ling-3.0-tiny

    inclusionAI/Ling-3.0-tiny

    2026-08 (day unknown)7.9B MoE · 1.3B active

    AA index11.1single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryKV unknown
    Unified16.2+ / 36.0 GiB
    Generate~20–57 t/sEstimatedextrapolated≈ meets 5 t/s
    First token · 4K prompt—Unknown
  5. 5

    diffusiongemma-26B-A4B-it

    google/diffusiongemma-26B-A4B-it

    2026-06-1025B MoE · 3.8B active

    AA index9.5 est.single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryKV unknown
    Unified26.5+ / 36.0 GiB
    Generate~13–37 t/sEstimatedextrapolated≈ meets 5 t/s
    First token · 4K prompt—Unknown
  6. 6

    gemma-4-E4B-it

    google/gemma-4-E4B-it

    2026-04-028B dense

    AA index8.9Reasoning
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryKV unknown
    Unified9.0+ / 36.0 GiB
    Generate~24–29 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~10 s–11 sEstimated
  7. 7

    gemma-4-E2B-it

    google/gemma-4-E2B-it

    2026-04-025.1B dense

    AA index7.8 est.Reasoning
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryKV unknown
    Unified6.1+ / 36.0 GiB
    Generate~39–47 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~6.7 s–7.0 sEstimated
  8. 8

    Phi-4-mini-instructOlder

    microsoft/Phi-4-mini-instruct

    2025-02-263.8B tensors · dense

    AA index6.3 est.single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryKV unknown
    Unified8.7+ / 36.0 GiB
    Generate~25–30 t/sEstimatedmeets 5 t/s
    First token · 4K prompt~5.0 s–5.3 sEstimated
  9. 9

    tiny-aya-globalGated

    CohereLabs/tiny-aya-global

    2026-02-173.3B tensors

    AA index4.8 est.single mode
    Plan & memory
    FP16llama.cpp MetalUnified memoryKV unknown
    Unified7.7+ / 36.0 GiB
    Generate—Unknownspeed unknown
    First token · 4K prompt—Unknown
+ 24 models without a AA index score

Speed unknown 5

No measurement, calibration or bandwidth figure for this plan.

  1. 1

    gpt-oss-20bOlder

    openai/gpt-oss-20b

    2025-08-0521B tensors · MoE

    AA index9.0High
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits
    Unified13.0 / 36.0 GiB
    Generate—Unknownspeed unknown
    First token · 4K prompt—Unknown
  2. 2

    granite-4.0-h-1b

    ibm-granite/granite-4.0-h-1b

    2025-10-281.5B tensors · MoE

    AA index5.2 est.single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits*
    Unified3.0 / 36.0 GiB
    Generate—Unknownspeed unknown
    First token · 4K prompt—Unknown
  3. 3

    granite-4.0-micro

    ibm-granite/granite-4.0-micro

    2025-10-023.4B tensors · MoE

    AA index5.1 est.single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits
    Unified5.5 / 36.0 GiB
    Generate—Unknownspeed unknown
    First token · 4K prompt—Unknown
  4. 4

    granite-4.0-350m

    ibm-granite/granite-4.0-350m

    2025-10-280.4B tensors · MoE

    AA index4.8 est.single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits
    Unified2.1 / 36.0 GiB
    Generate—Unknownspeed unknown
    First token · 4K prompt—Unknown
  5. 5

    granite-4.0-h-350m

    ibm-granite/granite-4.0-h-350m

    2025-10-280.3B tensors · MoE

    AA index4.8 est.single mode
    Plan & memory
    Q8_0llama.cpp MetalUnified memoryFits*
    Unified1.9 / 36.0 GiB
    Generate—Unknownspeed unknown
    First token · 4K prompt—Unknown
+ 4 models without a AA index score
104 models with no feasible plan on this setup

Specs & sources

Memory48 GB unifiedsource
Bandwidth273 GB/ssource
FP16unknownApple publishes no comparable FP16 throughput or GPU-only TDP in reviewed chip/device technical specifications. Power adapter rating and whole-system maximum continuous power are not GPU TDP.
PowerunknownApple publishes no comparable FP16 throughput or GPU-only TDP in reviewed chip/device technical specifications. Power adapter rating and whole-system maximum continuous power are not GPU TDP.
  • 64GB is supported by Mac mini; the MacBook Pro technical specifications only list 24/48GB for M4 Pro.

Found in: Mac mini (2024)

Reported speeds

15 observations

Resolved to this device (same chip and GPU-core count). Reported by the linked sources; not reproduced by LocalFit.

Open in speed databaseSubmit yours

Doesn't fit? Rent a GPU

  • RunpodPlaceholderCloud GPU pods you can rent for running larger models.
  • Vast.aiPlaceholderGPU rental marketplace with many hosts and GPU types.
  • LambdaPlaceholderCloud GPU instances for AI training and inference.
  • TensorDockPlaceholderGPU cloud marketplace with consumer and datacenter GPUs.

Sponsored placeholder links: LocalFit may earn a commission if these become affiliate links. They never influence fit results, estimates or picks.