LLMBOTTLENECK.COM
stepfun-ai / step3p5

Step-3.7-Flash

201.4 billion parameters.

Open in the calculator →

Architecturepublished data
Quick answer
Step-3.7-Flash needs about 128.2 GB at Q4_K_M — the format most people download — with 8,192 tokens of context. It fits entirely on 1 of 18 common devices, starting with the Apple M2 Ultra (192 GB). What else fits in 192 GB. No hardware for it? Rent a GPU that runs it, priced live.

Sized at 8,192 tokens of context with the whole model in device memory, across 18 common devices. Speeds are estimates, not benchmarks.

Find the Apple M2 Ultra, the smallest card listed here that holds it: Amazon ↗ · eBay (new and used) ↗Store links may pay us a commission. They never decide which card is suggested — the memory arithmetic does.

Same decoder-plus-MTP layer_types convention as Step-3.5-Flash. Licence declared on the repository: apache-2.0.

Weight-file size by format

FormatSizeRangeBasis
FP16402.9 GB281.9 GB – 406.8 GBsize range
Q8_0214.1 GB149.8 GB – 217.1 GBsize range
Q6_K165.4 GB115.6 GB – 168.1 GBsize range
Q5_K_M143.9 GB96.9 GB – 168.1 GBsize range
Q5_0140.5 GB96.9 GB – 168.1 GBsize range
Q4_K_M123.6 GB79.3 GB – 168.1 GBsize range
Q4_0117.2 GB79.3 GB – 168.1 GBsize range
Q3_K_M100.7 GB60.6 GB – 168.1 GBsize range
Q2_K79.7 GB46.3 GB – 168.1 GBsize range

A file size is not the memory a run needs: the KV cache and the runtime reserve come on top, and the calculator adds both.

What the labels mean
published data
Read from a published source — a file's byte count, a model's configuration or a manufacturer's specification — or exact arithmetic on such values. Not a measurement on a machine.
reconstructed size
Weight size reconstructed from the pinned architecture, because no published file exists.
size range
Only a lower and an upper bound are claimed for this weight size.

Hardware ladder

Q4_K_M at 8,192 tokens · 1 of 18 devices hold it entirely
DeviceMemoryVerdictDecodeCalculator
NVIDIA GeForce RTX 5070128.2 GB of 12 GB43 layers on system RAM~7.5 tok/s with offloadOpen →
AMD Radeon™ RX 9070 XT128.2 GB of 16 GB41 layers on system RAM~7.9 tok/s with offloadOpen →
NVIDIA GeForce RTX 4080128.2 GB of 16 GB41 layers on system RAM~7.8 tok/s with offloadOpen →
NVIDIA GeForce RTX 5080128.2 GB of 16 GB41 layers on system RAM~8.0 tok/s with offloadOpen →
NVIDIA GeForce RTX 5070 Ti128.2 GB of 16 GB41 layers on system RAM~8.0 tok/s with offloadOpen →
NVIDIA GeForce RTX 5060 Ti 16GB128.2 GB of 16 GB41 layers on system RAM~7.5 tok/s with offloadOpen →
NVIDIA GeForce RTX 4090128.2 GB of 24 GB38 layers on system RAM~8.6 tok/s with offloadOpen →
AMD Radeon™ RX 7900 XTX128.2 GB of 24 GB38 layers on system RAM~8.6 tok/s with offloadOpen →
NVIDIA GeForce RTX 3090128.2 GB of 24 GB38 layers on system RAM~8.5 tok/s with offloadOpen →
NVIDIA GeForce RTX 5090128.2 GB of 32 GB36 layers on system RAM~9.2 tok/s with offloadOpen →
NVIDIA RTX 6000 Ada Generation128.2 GB of 48 GB30 layers on system RAM~10 tok/s with offloadOpen →
Apple M4 Pro128.2 GB of 64 GBdoes not fitnot calibratedOpen →
Apple M5 Pro128.2 GB of 64 GBdoes not fitnot calibratedOpen →
AMD Ryzen AI Max+ 395 with Radeon 8060S128.2 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M3 Max128.2 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M4 Max128.2 GB of 128 GBdoes not fitnot calibratedOpen →
NVIDIA DGX Spark128.2 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M2 Ultra128.2 GB of 192 GBfits~47 tok/sOpen →

Decode is a calibrated estimate from the published calibration; capacity is the manufacturer's published ceiling, not guaranteed free memory. Fitting in memory is not the same as loading: whether the runtime and version you have supports this architecture and format on that machine has not been tested here. Offloaded rows assume there is enough system RAM for the overflow — the calculator checks that against the RAM you declare. Each “Open” link keeps this model, format and context.

Run it in the cloud

Rent a GPU that runs Step-3.7-Flash

Every rentable machine that holds the whole model, from Vast.ai and RunPod, with this site's speed estimate for each and the provider's own price, read live. Cheapest first; sort by cost per token or speed instead, or switch the format.

Prices · 18:30 UTC, 22 Sept
9 configurations in stock · 11 out of stock
MachineSpeedPer hourPer M tokensWhereRent
4× L40S192 GBcheapest~203 tok/sFaster than you read$1.87$2.56Vast.aiMarketplace · 99.2% reliableRent on Vast.ai
2× RTX PRO 6000192 GB~210 tok/sFaster than you read$2.27$2.99Vast.aiMarketplace · 98.8% reliableRent on Vast.ai
4× RTX 6000 Ada192 GB~225 tok/sFaster than you read$3.00$3.69Vast.aiMarketplace · 98.5% reliableRent on Vast.ai
4× RTX PRO 5000192 GB~315 tok/sFaster than you read$3.28$2.89RunPodCommunity CloudRent on RunPod
2× H100 SXM160 GB~393 tok/sFaster than you read$3.47$2.45Vast.aiMarketplace · 98.7% reliableRunPod $6.98/h Rent on Vast.ai
1× H200141 GB~282 tok/sFaster than you read$3.59$3.54RunPodCommunity CloudRent on RunPod
8× RTX 4090192 GBrecommendedbest valuefastest~473 tok/sFaster than you read$4.06$2.38Vast.aiMarketplace · 99.2% reliableRent on Vast.ai
2× H100 PCIe160 GB~235 tok/sFaster than you read$4.79$5.66Vast.aiMarketplace · 98.7% reliableRent on Vast.ai

No GPU at all

Use Step-3.7-Flash by the token

For one person chatting, that is about 2.1× cheaper than the best-value rented card above ($2.38 per million tokens). A rented GPU pays off when you keep it busy — many requests batched together, long agent runs — or when the data must stay on a machine you control, or the exact file you want is not served anywhere.

First time renting a GPU? How it works, in four steps
  1. Create an account and add credit. Both providers are prepaid: $10 is enough to try any card on this page for hours. A new RunPod account that arrives through a referral link gets a one-time $5 credit when it first adds $10.
  2. Pick the card and the count shown here, and a template: llama.cpp or vLLM for an OpenAI-compatible API, or one with a web chat if you only want to talk to the model. “How to launch it” below gives the exact image and command.
  3. Wait for the download. The weights are fetched on the machine; tens of gigabytes take a few minutes on a data-centre connection. You pay by the second from the moment it starts.
  4. Stop it when you are done. A running machine bills even when idle. A stopped one costs nothing per hour, though its disk is usually still billed until you delete it.
How to launch it

No Q4_K_M file of Step-3.7-Flash is catalogued here, so there is no exact command to vouch for. Start the machine from the llama.cpp server image ghcr.io/ggml-org/llama.cpp:server-cuda and point -hf at a Q4_K_M build from the publisher or a quantizer you trust — search Hugging Face for one. Check its file size against the memory figure on this page before you rent.

Every machine holds the whole model at Q4_K_M and 8,192 tokens of context, no offload. Speeds are this site’s single-stream decode estimates; prices are what each provider’s own API quoted, on-demand, for the whole machine. Vast hosts below 98% measured reliability are left out.

Referral links Vast.ai, RunPod and Novita pay us a share of what you spend if you sign up through these buttons. It costs you nothing, and it never decides an order or a recommendation: both are computed from the live price and the speed, and options that pay us nothing are listed and recommended on the same terms. How we rank

Buying a card instead, or paying by the token? Compare a year of Step-3.7-Flash three ways →

Under the hood

Architecture, read from the publisher’s file

The numbers every figure above is computed from, with the file they came from.

Architecture

✓ Architecture read from the published config.json

Retrieved 2026-09-20 at pinned commit 5f6244077ac6; parameter count from the safetensors index at the same revision.

Architecture
step3p5
Layers
45
Hidden size
4,096
Attention heads
64
Head dimension
128
Feed-forward width
11,264
Vocabulary
128,896
Context ceiling
262,144
Sliding window
512
Expert width
1,280

Where the memory goes

tokenembedding× 45 decoder blocksMulti-head attentionhead counts not publishedwindow 512Feed-forwardone networkall activeoutputprojectiongrows with contextfixed per token

Every attention head caches its own keys and values, so the KV cache grows at the full multi-head rate with context.

This is a multimodal checkpoint (vision); the diagram and memory sizing above cover the text decoder. Encoder/audio/image activations are not included in the KV or weight figures.

Open the exact official config.json ↗

More from stepfun-ai

Step-3.5-Flash

Use this answer in your own product

As JSON, from the API
curl -s https://llmbottleneck.com/v1/analyze \
  -H "Authorization: Bearer $LLMB_KEY" -H "content-type: application/json" \
  -d '{"model":"stepfun-ai-step-3-7-flash","quantization":"Q4_K_M","context":8192,"hardware":"nvidia-geforce-rtx-4090"}'

Same engine, same evidence, every field sourced. Free key in one step →

On your page, as a widget
<script src="https://llmbottleneck.com/widget.js"
  data-model="stepfun-ai-step-3-7-flash" data-quantization="Q4_K_M"
  data-hardware="nvidia-geforce-rtx-4090" data-context="8192"></script>

No key needed. Unbranded, with your own buy button, on Pro and Business →

Citing this page

LLM Bottleneck. “Step-3.7-Flash VRAM and hardware requirements.” Architecture from stepfun-ai/Step-3.7-Flash at revision 5f6244077ac6, retrieved 2026-09-20. https://llmbottleneck.com/models/stepfun-ai-step-3-7-flash

Every figure above is either the published value or a reconstruction whose measured error is on the accuracy page.

llmbottleneck
catalogue 2026-10-03models 327devices 135