VibeThinker-1.5B
by WeiboAI
1.8B params · text-generation · 545 likes · 1.7k downloads
VibeThinker-1.5B is a 1.8B parameter model. At Q4 quantization it requires 1GB of VRAM. It runs comfortably on GeForce RTX 4090 (738 tok/s), GeForce RTX 5090 (1107 tok/s), M4 Max 128GB (270 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Featherless |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 738 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 1107 tok/s | comfortable |
| M4 Max 128GB | 128GB | 270 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 135 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 135 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 1353 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 2722 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 653 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 540 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 524 tok/s | comfortable |