S5 nano
S5 nano is an open-weights language model from Maritaca-ai, released November 2, 2026. It has 4.7B parameters (dense). The context window is 256K tokens.
Check it against your own PC →
Can you run S5 nano locally?
The smallest common GPU it runs on is 6 GB of VRAM (with 32 GB of RAM): Q6_K, fully on GPU, roughly 29 tokens per second. Estimates assume 32 GB of DDR5 RAM, a 16K context and llama.cpp-style GGUF quantization. Enter your exact hardware in the live app.
| GPU memory | Verdict | Best quant | VRAM | RAM | tok/s |
|---|---|---|---|---|---|
| 6 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.0 GB | – | ~29 |
| 8 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.0 GB | – | ~41 |
| 12 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.0 GB | – | ~54 |
| 16 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.0 GB | – | ~73 |
| 24 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.0 GB | – | ~130 |
| 32 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.0 GB | – | ~220 |
Ready-made quantizations on Hugging Face: bartowski/maritaca-ai_S5-nano-GGUF · maritaca-ai/S5-nano-GGUF · NANI-Nithin/S5-nano-GGUF · NoxNotreve/maritaca-ai_S5-nano-GGUF · necropopeost/S5-nano-MLX-4bit-vision · necropopeost/S5-nano-oQ5e
Benchmark results
Epoch AI has not published results for S5 nano yet. New models are usually evaluated within days to weeks.