HyperCLOVA X SEED 32B Think
HyperCLOVA X SEED 32B Think is an open-weights language model from NAVER, released December 29, 2025. It has 32B parameters (dense).
Developed by Naver, South Korea’s leading AI research lab, this cutting-edge language model supports multimodal inputs and advanced reasoning.
Check it against your own PC →
Lab
NAVER
Released
December 29, 2025
Weights
Open
Parameters
32B dense
Context
–
Price / M tokens
–
Epoch ECI
not yet rated
Can you run HyperCLOVA X SEED 32B Think locally?
The smallest common GPU it runs on is 16 GB of VRAM (with 32 GB of RAM): Q3_K_M, split GPU / CPU, roughly 5 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 | 11.0 GB | ~3 |
| 8 GB VRAM | Very slow | IQ3_XXS (noticeable loss) | 7.2 GB | 9.0 GB | ~3 |
| 12 GB VRAM | Very slow | IQ3_XXS (noticeable loss) | 11.2 GB | 5.0 GB | ~5 |
| 16 GB VRAM | Runs, slowly | Q3_K_M (okay) | 15.2 GB | 4.4 GB | ~5 |
| 24 GB VRAM | Runs on GPU | IQ4_XS (good) | 21.0 GB | – | ~32 |
| 32 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 30.2 GB | – | ~39 |
Benchmark results
Epoch AI has not published results for HyperCLOVA X SEED 32B Think yet. New models are usually evaluated within days to weeks.