Skip to content

Step 3.7 Flash VRAM requirements

How much VRAM Step 3.7 Flash needs
4-bit weights: 114 GBWith 8K context: 118 GBWith 32K context: 127 GB8-bit, 8K context: 204 GB
Parameters
201B
Active per token
9.5B
Longest context
256K
Released
May 2026
License
Apache 2.0
Strongest at
Text and images

How much VRAM Step 3.7 Flash needs

Memory Step 3.7 Flash needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit375 GB379 GB388 GB424 GB
8-bit199 GB204 GB213 GB249 GB
6-bit154 GB158 GB167 GB203 GB
5-bit133 GB138 GB147 GB183 GB
4-bit114 GB118 GB127 GB163 GB
3-bit91.7 GB96.0 GB105 GB141 GB
2-bit70.3 GB74.7 GB83.7 GB120 GB
1.6-bit38.7 GB43.1 GB52.1 GB88.1 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Step 3.7 Flash

The best way Step 3.7 Flash runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Step 3.7 Flash on RTX 4060 report
RTX 5060Too bigPDF: Step 3.7 Flash on RTX 5060 report
RTX 3070Too bigPDF: Step 3.7 Flash on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Step 3.7 Flash on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Step 3.7 Flash on RTX 3060 12 GB report
Arc B580Too bigPDF: Step 3.7 Flash on Arc B580 report
RTX 4070Too bigPDF: Step 3.7 Flash on RTX 4070 report
RTX 4070 SuperToo bigPDF: Step 3.7 Flash on RTX 4070 Super report
RTX 5070Too bigPDF: Step 3.7 Flash on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Step 3.7 Flash on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Step 3.7 Flash on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Step 3.7 Flash on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Step 3.7 Flash on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Step 3.7 Flash on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Step 3.7 Flash on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Step 3.7 Flash on RTX 5070 Ti report
RTX 5080Too bigPDF: Step 3.7 Flash on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Step 3.7 Flash on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Step 3.7 Flash on RTX 3090 report
RTX 4090Too bigPDF: Step 3.7 Flash on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Step 3.7 Flash on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Step 3.7 Flash on RTX 5090 report
2× RTX 3090Too bigPDF: Step 3.7 Flash on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Step 3.7 Flash on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Step 3.7 Flash on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Step 3.7 Flash on RTX 6000 Ada report
L40SToo bigPDF: Step 3.7 Flash on L40S report
2× RTX 5090Too bigPDF: Step 3.7 Flash on 2× RTX 5090 report
4× RTX 30902-bit44K123 tok/s0.5 sPDF: Step 3.7 Flash on 4× RTX 3090 report
RTX PRO 6000 Blackwell2-bit48K180 tok/s0.5 sPDF: Step 3.7 Flash on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Step 3.7 Flash on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Step 3.7 Flash on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Step 3.7 Flash on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Step 3.7 Flash on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Step 3.7 Flash on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Step 3.7 Flash on Mac mini M4 Pro 64 GB report
M4 Max 128 GB2-bit52K67 tok/s9.9 sPDF: Step 3.7 Flash on M4 Max 128 GB report
M5 Max 128 GB2-bit52K71 tok/s2.9 sPDF: Step 3.7 Flash on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Step 3.7 Flash on M3 Ultra 96 GB report
M3 Ultra 256 GB6-bit72K48 tok/s5.6 sPDF: Step 3.7 Flash on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K40 tok/s5.6 sPDF: Step 3.7 Flash on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K52 tok/s1.7 sPDF: Step 3.7 Flash on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Step 3.7 Flash on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB2-bit52K37 tok/s5.7 sPDF: Step 3.7 Flash on Ryzen AI Max+ 395 128 GB report
RTX Spark3-bit56K40 tok/s2.0 sPDF: Step 3.7 Flash on RTX Spark report
DGX Spark3-bit56K37 tok/s2.1 sPDF: Step 3.7 Flash on DGX Spark report
Dual DGX Spark8-bit72K18 tok/s2.6 sPDF: Step 3.7 Flash on Dual DGX Spark report
DGX Station GB3008-bit96K211 tok/s0.2 sPDF: Step 3.7 Flash on DGX Station GB300 report
2× RTX PRO 6000 Blackwell6-bit64K112 tok/s0.3 sPDF: Step 3.7 Flash on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K93 tok/s0.2 sPDF: Step 3.7 Flash on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K93 tok/s0.1 sPDF: Step 3.7 Flash on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Step 3.7 Flash on A100 80 GB SXM report
H100 SXMToo bigPDF: Step 3.7 Flash on H100 SXM report
4× H200 NVL8-bit256K175 tok/s0.1 sPDF: Step 3.7 Flash on 4× H200 NVL report
8× H200 NVL8-bit256K175 tok/s0.1 sPDF: Step 3.7 Flash on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf unsloth/Step-3.7-Flash-GGUF
vLLM
vllm serve stepfun-ai/Step-3.7-Flash
MLX (Mac)
mlx_lm.chat --model mlx-community/Step-3.7-Flash-4bit

Questions

How much VRAM does Step 3.7 Flash need?

Step 3.7 Flash needs 114 GB for its weights at 4-bit, 118 GB with an 8K context and 127 GB with 32K. At 8-bit, 204 GB.

What is the minimum VRAM for Step 3.7 Flash?

About 74.7 GB at 2-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the 4× RTX 3090.

How big is Step 3.7 Flash?

201B parameters (9.5B active per token): 375 GB of weights at 16-bit, 114 GB at 4-bit.

Can Step 3.7 Flash run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the 4× RTX 3090.

More models