Mamba-3 1.5B
Mamba-3 1.5B is an open-weights language model from Carnegie Mellon University, released September 18, 2025. It has 1.5B parameters (dense).
The recent scaling of test-time compute for LLMs has restricted the practical deployment of models to those with strong capabilities that can generate high-quality outputs in an inference-efficient manner.
Check it against your own PC →
Lab
Carnegie Mellon University
Released
September 18, 2025
Weights
Open
Parameters
1.5B dense
Context
–
Price / M tokens
–
Epoch ECI
not yet rated
Can you run Mamba-3 1.5B locally?
The smallest common GPU it runs on is 6 GB of VRAM (with 32 GB of RAM): Q6_K, fully on GPU, roughly 86 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) | 3.4 GB | – | ~86 |
| 8 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 3.4 GB | – | ~120 |
| 12 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 3.4 GB | – | ~150 |
| 16 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 3.4 GB | – | ~200 |
| 24 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 3.4 GB | – | ~310 |
| 32 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 3.4 GB | – | ~470 |
Benchmark results
Epoch AI has not published results for Mamba-3 1.5B yet. New models are usually evaluated within days to weeks.