[ModelPage]: https://static.stepfun.com/blog/step-3.7-flash/

1. Introduction

GGUF quantizations of stepfun-ai/Step-3.7-Flash.

Step-3.7-Flash is a 198B-parameter sparse Mixture-of-Experts vision-language model from StepFun-ai, activating ~11B parameters per token for up to 400 t/s throughput. It pairs a 196B-parameter language backbone with a 1.8B-parameter vision encoder for native image understanding, supports a 256K context window, and offers three selectable reasoning levels (low / medium / high) to balance speed, cost, and depth. Built for agentic workloads — tool calling, multi-step reasoning, code, and math — with native multilingual coverage.

A separate mmproj projector ships alongside the language quants for multimodal inference. With 128 GB of unified memory (Mac Studio, DGX Spark, Ryzen AI Max+ 395, etc.), you can privately host Step-3.7-Flash: Q4 quants and below run at full 256K context with high precision.

2. Files

3. Quickstart

Build llama.cpp and run:

# 1. Clone and build
git clone https://github.com/stepfun-ai/llama.cpp.git
cd llama.cpp
git checkout -b step3.7 origin/step3.7
cmake -B build -DLLAMA_BUILD_TOOLS=ON -DLLAMA_BUILD_SERVER=ON
cmake --build build --config Release -j$(nproc)

# 2. Test performance (benchmark)
./build/bin/llama-batched-bench \
  -m Step-3.7-flash-Q4_K_S.gguf \
  -c 32768 -b 2048 -ub 2048 \
  -npp 0,2048,8192,16384,32768 -ntg 128 -npl 1

# 3. Text-only inference
./build/bin/llama-cli \
  -m Step-3.7-flash-Q4_K_S.gguf \
  -c 32768 -ngl 99 -fa on \
  -p "Write a Python function to compute the n-th Fibonacci number."

# 4. With vision (image + text)
./build/bin/llama-mtmd-cli \
  -m Step-3.7-flash-Q4_K_S.gguf \
  --mmproj mmproj-Step-3.7-flash-f16.gguf \
  -c 32768 -ngl 99 -fa on \
  --image path/to/image.jpg \
  -p "Describe this image."

# 5. OpenAI-compatible server (text + vision)
./build/bin/llama-server \
  -m Step-3.7-flash-Q4_K_S.gguf \
  --mmproj mmproj-Step-3.7-flash-f16.gguf \
  -c 32768 -ngl 99 -fa on \
  --host 0.0.0.0 --port 8080

For full CLI / server options, see the llama.cpp README.

4. Performance

Apple Mac Studio (M4 max, 128 GB unified memory)

Step-3.7-flash-Q4_K_S

./llama-batched-bench -m Step-3.7-flash-Q4_K_S.gguf -c 262150 -b 2048 -ub 1024 -npp 0,2048,8192,16384,32768,65536,131072,262144 -ntg 128 -npl 1

Step-3.7-flash-IQ4_XS

./llama-batched-bench -m Step-3.7-flash-IQ4_XS.gguf -c 262150 -b 2048 -ub 1024 -npp 0,2048,8192,16384,32768,65536,131072,262144 -ntg 128 -npl 1

Step-3.7-flash-Q3_K_L

./llama-batched-bench -m Step-3.7-flash-Q3_K_L.gguf -c 262272 -b 2048 -ub 1024 -npp 0,2048,8192,16384,32768,65536,131072,262144 -ntg 128 -npl 1

NVIDIA DGX Spark (GB10, 128 GB unified memory)

Step-3.7-flash-Q4_K_S

./llama-batched-bench -m Step-3.7-flash-Q4_K_S.gguf -c 131300 -b 2048 -ub 1024 -npp 0,2048,8192,16384,32768,65536,131072 -ntg 128 -npl 1

Step-3.7-flash-IQ4_XS

./llama-batched-bench -m Step-3.7-flash-IQ4_XS.gguf -c 262272 -b 2048 -ub 1024 -npp 0,2048,8192,16384,32768,65536,131072,262144 -ntg 128 -npl 1

Step-3.7-flash-Q3_K_L

./llama-batched-bench -m Step-3.7-flash-Q3_K_L.gguf -c 262272 -b 2048 -ub 1024 -npp 0,2048,8192,16384,32768,65536,131072,262144 -ntg 128 -npl 1

AMD Ryzen AI Max+ 395 (Strix Halo, 128 GB unified memory)

Step-3.7-flash-Q4_K_S

llama-batched-bench.exe -m Step-3.7-flash-Q4_K_S.gguf -c 65664 -b 2048 -ub 1024 -npp 0,2048,8192,16384,32768,65536 -ntg 128 -npl 1

Step-3.7-flash-IQ4_XS

./llama-batched-bench -m Step-3.7-flash-IQ4_XS.gguf -c 65664 -b 2048 -ub 1024 -npp 0,2048,8192,16384,32768,65536 -ntg 128 -npl 1

Step-3.7-flash-Q3_K_L

./llama-batched-bench -m Step-3.7-flash-Q3_K_L.gguf -c 262272 -b 2048 -ub 1024 -ctk q8_0 -ctv q8_0 -npp 0,2048,8192,16384,32768,65536,131072,262144 -ntg 128 -npl 1

5. Acknowledgments

This release stands on the work of the following authors and communities:

• bartowski — for calibration_datav5, the community-standard imatrix calibration anchor used by countless GGUF releases. Used for calibration purposes only; no license has been verified for this resource.
• eaddario — for the imatrix-calibration dataset (MIT), providing multilingual / code / math splits that form the backbone of this release's domain balance
• NousResearch — for hermes-function-calling-v1 (Apache-2.0), used for agent / tool-call calibration coverage
• ggml-org / llama.cpp — for the entire quantization and inference toolchain (MIT)

6. License

The GGUF quantization files in this repository are derivative works of stepfun-ai/Step-3.7-Flash and are released under the same Apache 2.0 license.

All calibration datasets retain their original licenses and are used strictly for quantization calibration purposes only.