Best Local LLMs for 48GB RAM

Ranked open-weight models that run well on a 48GB machine, with estimated speed and the run command for each.

Try: Qwen3 8B, RTX 4090, MacBook Pro M4, 16GB, iPhone 17 Pro

Ranked picks

Best local models for ~48GB

12 picks

Estimates assume a representative Apple-Silicon machine with 48GB unified memory. Tok/s are ModelFit estimates, not measured benchmarks; typical error is the median gap to measured runs on hardware the model was not fitted on (how the estimates work). Run the wizard for figures tuned to your exact chip.

  1. 01

    Qwen3.6 35B-A3B

    Qwen / 35B / Q4_K_M / ~22 GB

    Best for reasoning, coding, agents. Strong fit for 48 GB RAM with balanced speed and quality.

    ollama
    $ollama run qwen3.6:35b-a3b
    Fit
    Runs wellB · 37
    Best for
    Reasoning, Coding, Agents
    Perf
    est. 163 tok/s, typical error 10% · first token ~0.2s
  2. 02

    Nemotron 3.5 Lightning 30B-A3B

    Nemotron / 30B / Q4_K_M / ~25.4 GB

    Best for agentic, coding, long context. Strong fit for 48 GB RAM with balanced speed and quality.

    ollama
    $ollama run nemotron-3.5-lightning:30b
    Fit
    Runs wellB · 27
    Best for
    Agentic, Coding, Long context
    Perf
    est. 103 tok/s, typical error 10% · first token ~0.2s
  3. 03

    Sage-A1.5

    Sage / 35.95B / Q4_K_M / ~21.7 GB

    Best for scientific research agents, long-horizon tasks. Strong fit for 48 GB RAM with balanced speed and quality.

    llama.cpp
    $llama-server -hf InternScience/Sage-A1.5-Q4_K_M-GGUF:Q4_K_M
    Fit
    Runs wellB · 38
    Best for
    Scientific research agents, Long-horizon tasks
    Perf
    est. 163 tok/s, typical error 10% · first token ~0.2s
  4. 04

    Nemotron 3.5 Lightning 30B-A3B (IQ4_XS)

    Nemotron / 30B / IQ4_XS / ~18.9 GB

    Best for agentic, coding, long context. Strong fit for 48 GB RAM with balanced speed and quality.

    llama.cpp
    $llama-server -hf bartowski/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF:IQ4_XS
    Fit
    Runs wellB · 46
    Best for
    Agentic, Coding, Long context
    Perf
    est. 132 tok/s, typical error 10% · first token ~0.2s
  5. 05

    GLM-4.7-Flash 30B-A3B

    Zhipu / 30B / Q4_K_M / ~19 GB

    Best for coding, agentic tasks, reasoning. Strong fit for 48 GB RAM with balanced speed and quality.

    ollama
    $ollama run glm-4.7-flash
    Fit
    Runs wellB · 45
    Best for
    Coding, Agentic tasks, Reasoning
    Perf
    est. 116 tok/s, typical error 10% · first token ~0.2s
  6. 06

    Gemma 4 26B-A4B (Q8)

    Gemma / 26B / Q8_0 / ~28.1 GB

    This model may feel memory-heavy on 48 GB RAM, but it is still listed for balanced speed and quality.

    ollama
    $ollama run gemma4:26b-a4b-it-q8_0
    Fit
    Runs wellC · 19
    Best for
    Chat, Coding, Multimodal
    Perf
    est. 75 tok/s, typical error 10% · first token ~0.2s
  7. 07

    Ornith 1.0 35B

    Ornith / 35B / Q4_K_M / ~21.2 GB

    Best for agentic coding. Strong fit for 48 GB RAM with balanced speed and quality.

    ollama
    $ollama run ornith:35b
    Fit
    Runs wellB · 39
    Best for
    Agentic coding
    Perf
    est. 166 tok/s, typical error 10% · first token ~0.2s
  8. 08

    Flint 40B-A2.5B

    Flint / 39.8B / NVFP4 / ~22.9 GB

    Best for agentic coding, long context. Strong fit for 48 GB RAM with balanced speed and quality.

    ollama
    $ollama run flint:40b
    Fit
    Runs wellB · 34
    Best for
    Agentic coding, Long context
    Perf
    est. 175 tok/s, typical error 10% · first token ~0.2s
  9. 09

    Gemma 4 31B

    Gemma / 31B / Q4_K_M / ~20 GB

    Best for quality, coding, multimodal. Strong fit for 48 GB RAM with balanced speed and quality.

    ollama
    $ollama run gemma4:31b
    Fit
    Runs wellB · 43
    Best for
    Quality, Coding, Multimodal
    Perf
    est. 24 tok/s, typical error 10% · first token ~1.6s
  10. 10

    Laguna XS 2.1

    Laguna / 33B / Q4_K_M / ~20.3 GB

    Best for agentic coding, long-horizon tasks. Strong fit for 48 GB RAM with balanced speed and quality.

    ollama
    $ollama run laguna-xs-2.1:q4_K_M
    Fit
    Runs wellB · 42
    Best for
    Agentic coding, Long-horizon tasks
    Perf
    est. 153 tok/s, typical error 10% · first token ~0.2s
  11. 11

    Qwen3.8 27B (Q8)

    Qwen / 27B / Q8_0 / ~27.1 GB

    This model may feel memory-heavy on 48 GB RAM, but it is still listed for balanced speed and quality.

    ollama
    $ollama run qwen3.8:27b-q8_0
    Fit
    Runs wellC · 22
    Best for
    Coding, Agent, Vision, Long context
    Perf
    est. 18 tok/s, typical error 10% · first token ~1.4s
  12. 12

    Gemma 4 26B-A4B

    Gemma / 26B / Q4_K_M / ~16 GB

    Best for chat, coding, multimodal. Strong fit for 48 GB RAM with balanced speed and quality.

    ollama
    $ollama run gemma4:26b
    Fit
    Runs wellB · 54
    Best for
    Chat, Coding, Multimodal
    Perf
    est. 108 tok/s, typical error 10% · first token ~0.2s

Browse by RAM tier

Want an exact recommendation?

The wizard tunes picks and speed estimates to your exact device, chip, and RAM.

Open ModelFit Wizard Browse all devices
Best Local LLMs for 48GB RAM (2026) | ModelFit