GLM-4.5 vs Hy3-preview
GLM-4.5 (358.3B parameters) and Hy3-preview (298.8B parameters) side by side: the memory each needs at every precision, what it costs to run on a live GPU, and the context window, KV cache and license where they are published. Numbers are computed from the models' published specs; this page does not rank quality.
Side by side
| Fact | GLM-4.5 | Hy3-preview |
|---|---|---|
| Parameters | 358.3B | 298.8B |
| Architecture | Grouped-query attention; mixture of 160 experts, 8 active per token | Grouped-query attention; mixture of 192 experts, 8 active per token |
| Context length | 131,072 tokens | 262,144 tokens |
| License | – | – |
| Published precision | BF16 | BF16 |
| VRAM needed, As published | 801 GB | 668 GB |
| VRAM needed, FP8 | 400 GB | 334 GB |
| VRAM needed, INT4 | 200 GB | 167 GB |
| Cheapest live fit, As published | AMD MI300X × 5 · $13.14/hr | AMD MI300X × 4 · $10.52/hr |
| Cheapest live fit, FP8 | RTX PRO 6000 WS × 5 · $7.84/hr | AMD MI300X × 2 · $5.26/hr |
| Cheapest live fit, INT4 | RTX A6000 × 5 · $1.82/hr | AMD MI300X · $2.63/hr |
| KV cache per token (16-bit) | 368 KB | 320 KB |
| KV cache at 32k tokens | 11.5 GB | 10.0 GB |
| KV cache at 128k tokens | 46.0 GB | 40.0 GB |
VRAM is the weight size at each precision times a flat 1.2 overhead; see the methodology. The FP8 and INT4 rows need a quantized checkpoint or an engine that quantizes on load. The fit is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does. KV cache is for one sequence at 16-bit, computed from each model's config where the attention layout is known.
Which to pick
- Hy3-preview needs less VRAM at its published precision (668 GB against 801 GB), so it fits on a smaller GPU.
- Hy3-preview lists the longer context window (262,144 tokens against 131,072).
- Hy3-preview caches less per sequence at 32k tokens (10.0 GB against 11.5 GB), leaving more memory for batching.
- Hy3-preview has the cheaper live GPU fit at its published precision ($10.52/hr against $13.14/hr).
These follow only from the facts in the table above. Whether either model does your task well is a separate question this page does not answer.
Keep reading
- GLM-4.5: full VRAM table and live GPU fit
- Hy3-preview: full VRAM table and live GPU fit
- The GLM model series
- The Hunyuan model series
Other comparisons