What can you run with 32GB of RAM?

With 32GB RAM you can run 65 of the 101 open-source models we track — up to about 36B parameters at Q4 quantization. The most popular that fits is Qwen3.5-9B, needing 6.8 GB and generating around ~10 tok/s on typical hardware of this size.

Typical machines: MacBook Pro M4 Pro · Gaming desktops · Mac mini M4 Pro. After the operating system takes its share, about 30 GB is free for a model.

Having the memory is not the same as having the speed

32GB RAM decides whether a model loads. How fast it answers is decided by memory bandwidth — how quickly your machine can read the model, which it must do once for every word it writes. Machines with 32GB RAM span 90–307 GB/s, so the fastest is about 3.4× quicker than the slowest with identical capacity. The table quotes a DDR5-5600; pick yours below to see the difference. Why bandwidth and not the processor →

90 GB/sDesktop or laptop — DDR5-5600· 5,600 MT/s × 128-bit dual channel ÷ 8 = 89.6 GB/s
100 GB/sMac — M2· Apple published specification for the M2: 100 GB/s
100 GB/sMac — M3· Apple published specification for the M3: 100 GB/s
120 GB/sMac — M4· Apple published specification for the M4: 120 GB/s
128 GB/sEnthusiast desktop — DDR5-8000· 8,000 MT/s × 128-bit dual channel ÷ 8 = 128 GB/s
137 GB/sCurrent thin laptop — LPDDR5X-8533· 8,533 MT/s × 128-bit ÷ 8 = 136.5 GB/s (soldered, 2 channels)
150 GB/sMac — M3 Pro· Apple published specification for the M3 Pro: 150 GB/s (lower than the M2 Pro)
153 GB/sMac — M5· Apple published specification for the M5: 153 GB/s
273 GB/sMac — M4 Pro· Apple published specification for the M4 Pro: 273 GB/s
307 GB/sMac — M5 Pro· Apple published specification for the M5 Pro: up to 307 GB/s

Models that run on 32GB RAM in 2026

ModelSizeMemoryEst. speedDownloadsFeels like
granite-embedding-97m-multilingual-r2Ibm97M655 MBRuns well~555 tok/s237,787Fast
granite-embedding-311m-multilingual-r2Ibm312M748 MBRuns well~311 tok/s47,167Fast
LFM2.5-Encoder-230MLiquid Ai230M793 MBRuns well~422 tok/s13,450Fast
LFM2.5-230MLiquid Ai230M808 MBRuns well~381 tok/s54,986Fast
ERNIE-4.5-0.3B-PTBaidu AI361M853 MBRuns well~243 tok/s21,323Fast
nemotron-3.5-asr-streaming-0.6bNVIDIA638M1.1 GBRuns well~118 tok/s1,052,774Fast
Qwen3.5-0.8BAlibaba / Qwen873M1.3 GBRuns well~110 tok/s2,990,867Fast
LFM2.5-1.2B-InstructLiquid Ai1.2B1.4 GBRuns well~80 tok/s555,431Fast
LFM2.5-VL-1.6BLiquid Ai1.6B1.4 GBRuns well~80 tok/s78,655Fast
Nemotron-3-Embed-1B-BF16NVIDIA1.1B1.5 GBRuns well~85 tok/s467,453Fast
cohere-transcribe-arabic-07-2026Cohere2.1B1.9 GBRuns well~47 tok/s51,389Fast
Hy-MT2-1.8BTencent2B1.9 GBRuns well~52 tok/s49,645Fast
Qwen3.5-2BAlibaba / Qwen2.3B2.0 GBRuns well~46 tok/s2,468,876Fast
granite-speech-4.1-2bIbm2.3B2.4 GBRuns well~37 tok/s402,320Fast
LFM2.5-2.6BLiquid Ai2.7B2.5 GBRuns well~35 tok/s77,973Fast
LFM2.5-Audio-1.5BLiquid Ai1.5B2.6 GBRuns well~32 tok/s1,025Fast
Unlimited-OCRBaidu AI3.3B2.7 GBRuns well~30 tok/s2,836,694Fast
LocateAnything-3BNVIDIA3.8B2.8 GBRuns well~28 tok/s463,853Fast
granite-4.1-3bIbm3.4B3.0 GBRuns well~28 tok/s300,243Fast
Shieldstral-1.0-3BMistral AI3.8B3.1 GBRuns well~27 tok/s2,480Fast
Ministral-3-3B-Instruct-2512Mistral AI3.8B3.1 GBRuns well~27 tok/s544,419Fast
Nemotron-Labs-Diffusion-3B-BaseNVIDIA3.8B3.3 GBRuns well~25 tok/s35,110Fast
Cosmos3-EdgeNVIDIA3.9B3.3 GBRuns well~25 tok/s53,138Fast
Nemotron-3.5-Content-SafetyNVIDIA4.3B3.6 GBRuns well~23 tok/s176,924Comfortable
gemma-4-E2B-itGoogle DeepMind5.1B3.8 GBRuns well~19 tok/s4,002,947Comfortable
Qwen3.5-4BAlibaba / Qwen4.7B3.8 GBRuns well~21 tok/s6,199,203Comfortable
Fara1.5-4BMicrosoft4.5B4.0 GBRuns well~20 tok/s4,882Comfortable
NVIDIA-Nemotron-3-Nano-4B-BF16NVIDIA4B4.1 GBRuns well~21 tok/s712,701Comfortable
Fara-7BMicrosoft8.3B5.5 GBRuns well~12 tok/s2,192Comfortable
Hy-MT2-7BTencent8B5.7 GBRuns well~13 tok/s26,629Comfortable
LFM2.5-8B-A1BLiquid Ai · MoE8.5B5.9 GBRuns well~96 tok/s1B active170,464Fast
Nemotron-3-Embed-8B-BF16NVIDIA8B5.9 GBRuns well~12 tok/s142,381Comfortable
gemma-4-E4B-itGoogle DeepMind8B6.2 GBRuns well~11 tok/s5,295,850Comfortable
Nemotron-Labs-Diffusion-8B-BaseNVIDIA8.5B6.2 GBRuns well~11 tok/s413,039Comfortable
Ministral-3-8B-Instruct-2512Mistral AI8.9B6.3 GBRuns well~11 tok/s119,379Comfortable
granite-4.1-8bIbm8.8B6.3 GBRuns well~11 tok/s4,213,980Comfortable
Ornith-1.0-9BDeepreinforce9B6.7 GBRuns well~10 tok/s2,250,422Comfortable
Qwen3.5-9BAlibaba / Qwen9.7B6.8 GBRuns well~10 tok/s11,767,971Comfortable
GLM-4.6V-FlashZhipu AI10B7.0 GBRuns well~9.5 tok/s282,562Comfortable
Fara1.5-9BMicrosoft9.4B7.0 GBRuns well~9.9 tok/s14,526Comfortable
Olmo-3-7B-InstructAi27.3B7.2 GBRuns well~13 tok/s406,776Comfortable
gemma-4-12B-itGoogle DeepMind12B9.4 GBRuns well~8.2 tok/s2,963,990Comfortable
Ministral-3-14B-Instruct-2512Mistral AI14B9.7 GBRuns well~7.1 tok/s198,169Slow
Nemotron-Labs-Diffusion-14BNVIDIA14B9.7 GBRuns well~7.1 tok/s6,554Slow
Cosmos3-NanoNVIDIA16B11 GBRuns well~6.2 tok/s313,164Slow
ERNIE-4.5-21B-A3B-PTBaidu AI · MoE22B15 GBRuns well~32 tok/s3B active44,971Fast
Voxtral-Small-24B-2507Mistral AI24B16 GBRuns well~4.1 tok/s262,513Slow
Magistral-Small-2506Mistral AI24B16 GBRuns well~4.1 tok/s49,182Slow
Qwen3.5-27BAlibaba / Qwen28B19 GBRuns well~3.5 tok/s2,637,888Slow
diffusiongemma-26B-A4B-itGoogle DeepMind · MoE26B19 GBRuns well~22 tok/s4B active1,952,156Comfortable
gemma-4-26B-A4B-itGoogle DeepMind · MoE27B20 GBRuns well~23 tok/s4B active10,976,594Comfortable
ERNIE-4.5-VL-28B-A3B-PTBaidu AI · MoE29B20 GBRuns well~32 tok/s3B active218,879Fast
Qwen3.6-27BAlibaba / Qwen28B20 GBRuns well~3.4 tok/s6,572,759Slow
granite-4.1-30bIbm29B20 GBRuns well~3.3 tok/s20,228Slow
Fara1.5-27BMicrosoft27B20 GBRuns well~3.3 tok/s3,208Slow
Hy-MT2-30B-A3BTencent · MoE30B20 GBRuns well~32 tok/s3B active13,753Fast
North-Mini-Code-1.0Cohere · MoE30B22 GBRuns well~15 tok/s6.1B active23,720Comfortable
diffusiongemma-26B-A4B-it-NVFP4NVIDIA · MoE14B22 GBRuns well~11 tok/s4B active1,508,724Comfortable
GLM-4.7-FlashZhipu AI31B22 GBRuns well~3.1 tok/s2,071,716Slow
Olmo-3.1-32B-InstructAi232B23 GBRuns well~3.0 tok/s51,153Slow
Ornith-1.0-35BDeepreinforce · MoE35B24 GBRuns well~14 tok/s7B active2,736,644Comfortable
gemma-4-31B-itGoogle DeepMind31B24 GBRuns well~3.2 tok/s11,028,273Slow
Qwen3.5-35B-A3BAlibaba / Qwen · MoE36B25 GBRuns well~32 tok/s3B active2,539,444Fast
Qwen3.6-35B-A3BAlibaba / Qwen · MoE36B25 GBRuns well~32 tok/s3B active5,939,238Fast
Qwen-AgentWorld-35B-A3BAlibaba / Qwen · MoE35B25 GBRuns well~31 tok/s3B active84,349Fast

Sorted by memory, smallest first — click any column to change it.

Upgrade

Including Qwen3-Next-80B-A3B-Instruct (81B), NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-NVFP4 (45B) and Cosmos3-Super (65B) — around ~24 tok/s on a typical machine that size.

See everything 64GB RAM runs →

All figures at Q4_K_M quantization, 4K context. “Memory” includes the model, its context cache and runtime overhead. Speed is calculated, not benchmarked: 90 GB/s × 0.65 ÷ bytes read per token, assuming a DDR5-5600. Scale it by your own machine's bandwidth from the list above. Mixture-of-experts models read only their active parameters per token, which is why some large ones outrun smaller dense models.

Image generators that also fit

These make pictures rather than text. Their memory use scales with the resolution you generate at, not conversation length, so treat these figures as a floor rather than a ceiling.

What you'd need more memory for

The next models up, and what they ask for.

Other hardware

Verified against HuggingFace on 2026-08-08New to this? Start here →Runs on a laptop and up · last 12 months only