What can you run with 16GB of RAM?
With 16GB RAM you can run 45 of the 101 open-source models we track — up to about 16B 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 Air M4 · Mid-range Windows laptops · Mac mini (base). After the operating system takes its share, about 14 GB is free for a model.
Having the memory is not the same as having the speed
16GB 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 16GB RAM span 51–153 GB/s, so the fastest is about 3.0× 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 16GB RAM in 2026
| Model↕ | Size↕ | Memory↑ | Est. speed↕ | Downloads↕ | Feels like |
|---|---|---|---|---|---|
| granite-embedding-97m-multilingual-r2Ibm | 97M | 655 MBRuns well | ~555 tok/s | 237,787 | Fast |
| granite-embedding-311m-multilingual-r2Ibm | 312M | 748 MBRuns well | ~311 tok/s | 47,167 | Fast |
| LFM2.5-Encoder-230MLiquid Ai | 230M | 793 MBRuns well | ~422 tok/s | 13,450 | Fast |
| LFM2.5-230MLiquid Ai | 230M | 808 MBRuns well | ~381 tok/s | 54,986 | Fast |
| ERNIE-4.5-0.3B-PTBaidu AI | 361M | 853 MBRuns well | ~243 tok/s | 21,323 | Fast |
| nemotron-3.5-asr-streaming-0.6bNVIDIA | 638M | 1.1 GBRuns well | ~118 tok/s | 1,052,774 | Fast |
| Qwen3.5-0.8BAlibaba / Qwen | 873M | 1.3 GBRuns well | ~110 tok/s | 2,990,867 | Fast |
| LFM2.5-1.2B-InstructLiquid Ai | 1.2B | 1.4 GBRuns well | ~80 tok/s | 555,431 | Fast |
| LFM2.5-VL-1.6BLiquid Ai | 1.6B | 1.4 GBRuns well | ~80 tok/s | 78,655 | Fast |
| Nemotron-3-Embed-1B-BF16NVIDIA | 1.1B | 1.5 GBRuns well | ~85 tok/s | 467,453 | Fast |
| cohere-transcribe-arabic-07-2026Cohere | 2.1B | 1.9 GBRuns well | ~47 tok/s | 51,389 | Fast |
| Hy-MT2-1.8BTencent | 2B | 1.9 GBRuns well | ~52 tok/s | 49,645 | Fast |
| Qwen3.5-2BAlibaba / Qwen | 2.3B | 2.0 GBRuns well | ~46 tok/s | 2,468,876 | Fast |
| granite-speech-4.1-2bIbm | 2.3B | 2.4 GBRuns well | ~37 tok/s | 402,320 | Fast |
| LFM2.5-2.6BLiquid Ai | 2.7B | 2.5 GBRuns well | ~35 tok/s | 77,973 | Fast |
| LFM2.5-Audio-1.5BLiquid Ai | 1.5B | 2.6 GBRuns well | ~32 tok/s | 1,025 | Fast |
| Unlimited-OCRBaidu AI | 3.3B | 2.7 GBRuns well | ~30 tok/s | 2,836,694 | Fast |
| LocateAnything-3BNVIDIA | 3.8B | 2.8 GBRuns well | ~28 tok/s | 463,853 | Fast |
| granite-4.1-3bIbm | 3.4B | 3.0 GBRuns well | ~28 tok/s | 300,243 | Fast |
| Shieldstral-1.0-3BMistral AI | 3.8B | 3.1 GBRuns well | ~27 tok/s | 2,480 | Fast |
| Ministral-3-3B-Instruct-2512Mistral AI | 3.8B | 3.1 GBRuns well | ~27 tok/s | 544,419 | Fast |
| Nemotron-Labs-Diffusion-3B-BaseNVIDIA | 3.8B | 3.3 GBRuns well | ~25 tok/s | 35,110 | Fast |
| Cosmos3-EdgeNVIDIA | 3.9B | 3.3 GBRuns well | ~25 tok/s | 53,138 | Fast |
| Nemotron-3.5-Content-SafetyNVIDIA | 4.3B | 3.6 GBRuns well | ~23 tok/s | 176,924 | Comfortable |
| gemma-4-E2B-itGoogle DeepMind | 5.1B | 3.8 GBRuns well | ~19 tok/s | 4,002,947 | Comfortable |
| Qwen3.5-4BAlibaba / Qwen | 4.7B | 3.8 GBRuns well | ~21 tok/s | 6,199,203 | Comfortable |
| Fara1.5-4BMicrosoft | 4.5B | 4.0 GBRuns well | ~20 tok/s | 4,882 | Comfortable |
| NVIDIA-Nemotron-3-Nano-4B-BF16NVIDIA | 4B | 4.1 GBRuns well | ~21 tok/s | 712,701 | Comfortable |
| Fara-7BMicrosoft | 8.3B | 5.5 GBRuns well | ~12 tok/s | 2,192 | Comfortable |
| Hy-MT2-7BTencent | 8B | 5.7 GBRuns well | ~13 tok/s | 26,629 | Comfortable |
| LFM2.5-8B-A1BLiquid Ai · MoE | 8.5B | 5.9 GBRuns well | ~96 tok/s1B active | 170,464 | Fast |
| Nemotron-3-Embed-8B-BF16NVIDIA | 8B | 5.9 GBRuns well | ~12 tok/s | 142,381 | Comfortable |
| gemma-4-E4B-itGoogle DeepMind | 8B | 6.2 GBRuns well | ~11 tok/s | 5,295,850 | Comfortable |
| Nemotron-Labs-Diffusion-8B-BaseNVIDIA | 8.5B | 6.2 GBRuns well | ~11 tok/s | 413,039 | Comfortable |
| Ministral-3-8B-Instruct-2512Mistral AI | 8.9B | 6.3 GBRuns well | ~11 tok/s | 119,379 | Comfortable |
| granite-4.1-8bIbm | 8.8B | 6.3 GBRuns well | ~11 tok/s | 4,213,980 | Comfortable |
| Ornith-1.0-9BDeepreinforce | 9B | 6.7 GBRuns well | ~10 tok/s | 2,250,422 | Comfortable |
| Qwen3.5-9BAlibaba / Qwen | 9.7B | 6.8 GBRuns well | ~10 tok/s | 11,767,971 | Comfortable |
| GLM-4.6V-FlashZhipu AI | 10B | 7.0 GBRuns well | ~9.5 tok/s | 282,562 | Comfortable |
| Fara1.5-9BMicrosoft | 9.4B | 7.0 GBRuns well | ~9.9 tok/s | 14,526 | Comfortable |
| Olmo-3-7B-InstructAi2 | 7.3B | 7.2 GBRuns well | ~13 tok/s | 406,776 | Comfortable |
| gemma-4-12B-itGoogle DeepMind | 12B | 9.4 GBRuns well | ~8.2 tok/s | 2,963,990 | Comfortable |
| Ministral-3-14B-Instruct-2512Mistral AI | 14B | 9.7 GBRuns well | ~7.1 tok/s | 198,169 | Slow |
| Nemotron-Labs-Diffusion-14BNVIDIA | 14B | 9.7 GBRuns well | ~7.1 tok/s | 6,554 | Slow |
| Cosmos3-NanoNVIDIA | 16B | 11 GBRuns well | ~6.2 tok/s | 313,164 | Slow |
Sorted by memory, smallest first — click any column to change it.
Step up to 24GB RAM and 11 more models come within reach
Including gemma-4-26B-A4B-it (27B), Qwen3.6-27B (28B) and Qwen3.5-27B (28B) — around ~31 tok/s on a typical machine that size.
See everything 24GB 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.