What can you run with 32GB of RAM?
With 32GB RAM you can run 22 of the 51 open-source models we track — up to about 36B parameters at Q4 quantization. The most popular that fits is Qwen3.8-27B, needing 19 GB and generating around ~3.6 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 →
Models that run on 32GB RAM in 2026
| Model↕ | Size↕ | Memory↑ | Est. speed↕ | Downloads↕ | Feels like |
|---|---|---|---|---|---|
| LFM2.5-230MLiquid Ai | 230M | 808 MBRuns well | ~381 tok/s | 58,011 | Fast |
| MiniCPM5-1BOpenBMB | 1.1B | 1.3 GBRuns well | ~90 tok/s | 667,154 | Fast |
| MiniCPM5-2BOpenBMB | 2.5B | 2.3 GBRuns well | ~37 tok/s | 206,774 | Fast |
| LFM2.5-2.6BLiquid Ai | 2.7B | 2.5 GBRuns well | ~35 tok/s | 98,713 | Fast |
| LFM2.5-VL-3BLiquid Ai | 3.1B | 2.5 GBRuns well | ~35 tok/s | 31,296 | Fast |
| North-Micro-Vision-InstructCohere | 2.5B | 2.5 GBRuns well | ~39 tok/s | 120,354 | Fast |
| granite-4.2-3bIbm | 3.7B | 3.1 GBRuns well | ~26 tok/s | 32,017 | Fast |
| Mage-VLMicrosoft | 4.7B | 3.9 GBRuns well | ~22 tok/s | 29,690 | Comfortable |
| LFM2.5-8B-A1BLiquid Ai · MoE | 8.5B | 5.9 GBRuns well | ~96 tok/s1B active | 80,632 | Fast |
| Ling-3.0-tinyInclusionai · MoE | 7.9B | 6.2 GBRuns well | ~61 tok/s1.6B active | 25,727 | Fast |
| granite-4.2-8bIbm | 8.8B | 6.6 GBRuns well | ~11 tok/s | 52,946 | Comfortable |
| Ornith-1.5-9BDeepreinforce | 9.7B | 6.9 GBRuns well | ~10 tok/s | 453,339 | Comfortable |
| gemma-4-12B-itGoogle DeepMind | 12B | 9.6 GBRuns well | ~8.1 tok/s | 2,958,440 | Comfortable |
| Qwen3.8-27BAlibaba / Qwen | 28B | 19 GBRuns well | ~3.6 tok/s | 7,703,400 | Slow |
| diffusiongemma-26B-A4B-itGoogle DeepMind · MoE | 26B | 19 GBRuns well | ~22 tok/s4B active | 708,393 | Comfortable |
| Muse-Glimmer-30BMeta AI | 30B | 20 GBRuns well | ~3.3 tok/s | 441,881 | Slow |
| granite-4.2-30bIbm | 29B | 21 GBRuns well | ~3.3 tok/s | 17,894 | Slow |
| North-Mini-Code-1.0Cohere · MoE | 30B | 22 GBRuns well | ~15 tok/s6.1B active | 11,742 | Comfortable |
| Laguna-XS-2.1Poolside · MoE | 33B | 23 GBRuns well | ~14 tok/s6.7B active | 32,445 | Comfortable |
| Nex-N2.5-miniNex Agi · MoE | 35B | 24 GBRuns well | ~14 tok/s7B active | 4,543 | Comfortable |
| Ornith-1.5-35B-A3BDeepreinforce · MoE | 36B | 24 GBRuns well | ~32 tok/s3B active | 359,808 | Fast |
| NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4NVIDIA · MoE | 18B | 28 GBTight | ~14 tok/s3B active | 1,261,605 | Comfortable |
Sorted by memory, smallest first — click any column to change it.
Step up to 64GB RAM and 2 more models come within reach
Including MiniMax-H3 (33B) and NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-NVFP4 (45B) — around ~1.1 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.
What you'd need more memory for
The next models up, and what they ask for.