Step-3.7 needs about 138.5 GB of memory at its main file (a 123.2 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 7 quants, from 72.9 GB to 140.9 GB.
Step-3.7 Flash – 198B‑parameter vision‑language MoE model
Step-3.7 Flash is a 198B‑parameter sparse Mixture‑of‑Experts vision‑language model. It combines a 196B‑parameter language backbone with a 1.8B‑parameter vision encoder. The model supports a 256k context window and activates about 11B parameters per token.
Good at
- High‑throughput multimodal reasoning
- Large‑context text and image processing
- Agentic workflow orchestration
Limits
- Requires large VRAM (≈82–158 GB at 8 k context)
- No explicit language list provided
- Licensed under Apache‑2.0
Best for
- Parsing large financial reports
- Running multi‑step search loops
- Concurrent coding agents
Who it’s for. Developers building agentic workflows that combine perception, search, and reasoning. Suitable for production workloads needing native image understanding and high‑throughput pipelines.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 72.9 GB | 82.2 GB | Too big | Get file → | |
| — | 90.3 GB | 101.7 GB | Too big | Get file → | |
| — | 90.3 GB | 101.7 GB | Too big | Get file → | |
| — | 123.2 GB | 138.5 GB | Too big | Get file → | |
| — | 123.2 GB | 138.5 GB | Too big | Get file → | |
| — | 140.9 GB | 158.3 GB | Too big | Get file → | |
| — | 140.9 GB | 158.3 GB | Too big | Get file → |
For 8 GB VRAM use the smallest file (72.92 GB, 82.2 GB memory). For 16 GB VRAM use the 90.32 GB file (101.7 GB memory). For 32 GB VRAM use the 123.24 GB file (138.5 GB memory). Estimates are approximate.
What GPU do you need for Step-3.7?
These figures add the file size, an fp16 KV cache at an 8K context window and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: no quant fits at an 8K context.
- 32 GB VRAM: no quant fits at an 8K context.
- 48 GB VRAM: no quant fits at an 8K context.
Which Step-3.7 quant should you choose?
If memory is tight, the smallest file is the first one in the table at 72.9 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Step-3.7
How much VRAM does Step-3.7 need?
About 138.5 GB at its main file with an 8K context window. That is the 123.2 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run Step-3.7 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 82.2 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Step-3.7 GGUF?
It depends on the quant: file is 72.9 GB, file is 90.3 GB, file is 90.3 GB, file is 123.2 GB, file is 123.2 GB, file is 140.9 GB and 1 more.
What context length does Step-3.7 support?
The GGUF header lists a trained context length of 262,144 tokens (256K).
Where can I download Step-3.7 GGUF?
The files are on Hugging Face in the mudler/Step-3.7-Flash-APEX-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Step-3.7 use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run Step-3.7 Flash on 8 GB VRAM?
The smallest GGUF file requires about 82.2 GB of VRAM at 8 k context, so 8 GB is insufficient.
What license does Step-3.7 Flash use?
The model is licensed under Apache‑2.0.
Does Step-3.7 Flash support vision tasks?
Yes, it includes a 1.8B‑parameter vision encoder for native image understanding.