Aya Vision 32B
Aya Vision 32B is an open-weights language model from Cohere, released March 4, 2025. It has 33B parameters (dense). The context window is 16K tokens.
Check it against your own PC →
Can you run Aya Vision 32B locally?
The smallest common GPU it runs on is 16 GB of VRAM (with 32 GB of RAM): IQ3_XXS, split GPU / CPU, roughly 11 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 | Very slow | IQ3_XXS (noticeable loss) | 5.2 GB | 11.5 GB | ~3 |
| 8 GB VRAM | Very slow | IQ3_XXS (noticeable loss) | 7.2 GB | 9.5 GB | ~3 |
| 12 GB VRAM | Very slow | IQ3_XXS (noticeable loss) | 11.2 GB | 5.5 GB | ~5 |
| 16 GB VRAM | Runs, slowly | IQ3_XXS (noticeable loss) | 15.2 GB | 1.5 GB | ~11 |
| 24 GB VRAM | Runs on GPU | IQ4_XS (good) | 21.6 GB | – | ~31 |
| 32 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 31.2 GB | – | ~38 |
Ready-made quantizations on Hugging Face: unsloth/aya-vision-32b-unsloth-bnb-4bit · unsloth/aya-vision-32b-bnb-4bit
Benchmark results
Epoch AI has not published results for Aya Vision 32B yet. New models are usually evaluated within days to weeks.
Links
Hugging Face: CohereLabs/aya-vision-32b
More from Cohere
tiny aya en thinker · tiny aya l2 thinker · North Micro Vision Instruct · North Mini Code · Command A Plus · Tiny Aya Global · Tiny Aya Fire · Tiny Aya Water