Step 3.5 Flash Base — Hardware Requirements & GPU Compatibility
Chat
Step 3.5 Flash Base is a 197.8B-parameter open language model from StepFun in the Step 3 family. It supports a context window of up to 32,768 tokens. At Q4_K_M it needs about 130.52 GB of VRAM — see which GPUs and Macs can run it below.
Step 3.5 Flash Base runs on a 192 GB+ Mac or a datacenter GPU
Best local pick → Mac Pro M2 Ultra (192 GB) · Q4_K_M · 136.8 GB · ~53 tok/s
Macs from 128 GB (Q3_K_M · ~59 tok/s)
Apache 2.0
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Step 3.5 Flash Base
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve stepfun-ai/Step-3.5-Flash-Base --max-model-len 16384 --tensor-parallel-size 8vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on B200 (192 GB) (55 GB headroom) · Q4_K_M
- Generation speed
- ~283 tok/s
- generation speed
- Cost per 1M output tokens
- $5.88
- per 1M output tokens
How Much VRAM Does Step 3.5 Flash Base Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | Runs on | |
|---|---|---|---|---|
| 92.5 GB | 106 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Runs on Instinct MI250X 128 GB | |
| 106.0 GB | 119.6 GB | 3-bit medium — 3-bit medium quantization | Runs on Instinct MI250X 128 GB | |
| 130.5 GB | 144 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Runs on H200 NVL 141 GB | |
| 155 GB | 168.5 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Runs on B200 192 GB | |
| 179.5 GB | 193 GB | 6-bit, very good — 6-bit quantization, very good quality | Runs on B200 192 GB | |
| 217.5 GB | 231 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Runs on Instinct MI325X 256 GB | |
| 435.1 GB | 448.6 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Runs on 7× H100 SXM 80 GB |
Selected Q4_K_M
4-bit medium quantization136.8GBat 16K context
- Weights
- 129.3 GB
- KV cache
- 7.2 GB
- Overhead
- 0.3 GB
Server-class: ~136.8 GB at 16K context · one H200 NVL 141 GB
Which Hardware Can Run Step 3.5 Flash Base?
Q4_K_M · 130.5 GBStep 3.5 Flash Base (Q4_K_M) requires 130.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 170+ GB is recommended. No consumer GPU has enough memory. 4 devices with unified memory can run Step 3.5 Flash Base, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Pro M2 Ultra (192 GB).
128–144 GB
Q4_K_M · best fit Q3_K_M (112.4 GB)
4 GPUs · 2 Mac classes · 2 device families
GH200 Grace Hopper Superchip~257Simulate output on GH200 Grace Hopper SuperchipH200 NVL~255Simulate output on H200 NVLH200 SXM~255Simulate output on H200 SXMM2 Ultra*128 GB~59Simulate output on Mac Pro M2 Ultra (128 GB)M5 Max128 GB~51Simulate output on Mac Studio M5 Max (128 GB)GB10~36Simulate output on NVIDIA DGX SparkJetson T5000 (Thor)~36Simulate output on NVIDIA Jetson AGX Thor Developer Kit192+ GB
fits · 55.2 GB free · up to 32K ctx
6 GPUs · 1 Mac class · 2 device families
B200~283Simulate output on B200B300~283Simulate output on B300Instinct MI350X~99Simulate output on Instinct MI350XM3 Ultra*256–512 GB~54Simulate output on Mac Studio (M3 Ultra, 256GB)8x H100 SXM5~358Simulate output on NVIDIA DGX H1008x A100 SXM4~334Simulate output on NVIDIA DGX A100 640GB
By VRAM tier · typical GPUs and Macs at Q4_K_M
Consumer GPUs (up to 32 GB) and Macs up to 96 GB: too small for this model (needs ≥ 98.8 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →
Alternatives to Step 3.5 Flash Base
Step 3 family · 6 models →Specifications
- Publisher
- StepFun
- Family
- Step 3
- Parameters
- 197.8B
- Active parameters
- 8.4B per token
- Architecture
- Step3p5ForCausalLM
- Context Length
- 32,768 tokens
- Vocabulary Size
- 128,815
- Release Date
- 2026-03-02
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Step 3.5 Flash Base need?
Step 3.5 Flash Base requires 130.5 GB of VRAM at Q4_K_M, or 435.1 GB at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 197.8B × 4.8 bits ÷ 8 = 118.7 GB
KV Cache + Overhead ≈ 11.8 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
VRAM usage by quantization
Can NVIDIA GeForce RTX 5090 run Step 3.5 Flash Base?
No — Step 3.5 Flash Base requires at least 92.5 GB at Q2_K, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
What's the best quantization for Step 3.5 Flash Base?
For Step 3.5 Flash Base, Q4_K_M (130.5 GB) offers the best balance of quality and VRAM usage. Q5_K_M (155 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 92.5 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Step 3.5 Flash Base on a Mac?
Yes — Mac Pro M2 Ultra (128 GB) and 14 other Macs can run Step 3.5 Flash Base, the smallest at Q3_K_M. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 130.5 GB of usable unified memory (RAM minus macOS overhead), from Mac Pro M2 Ultra (192 GB).
Can I run Step 3.5 Flash Base locally?
Yes — Step 3.5 Flash Base can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Step 3.5 Flash Base?
At Q4_K_M, Step 3.5 Flash Base can reach ~99 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 5.5 GB active ÷ (8000 × 0.65) = 1.06510 ms, plus 45 layers × 0.055 ms = 2.47500 ms, so 1000 ÷ 3.54010 ms = ~282 tok/s
Estimated speed at Q4_K_M (130.5 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Step 3.5 Flash Base?
At Q4_K_M, the download is about 118.66 GB. The full-precision BF16 version is 395.53 GB. The smallest option (Q2_K) is 84.05 GB.
Which GPUs can run Step 3.5 Flash Base?
No single consumer GPU has enough VRAM to run Step 3.5 Flash Base at Q4_K_M (130.5 GB). Multi-GPU or professional hardware is required.
Which devices can run Step 3.5 Flash Base?
4 devices with unified memory can run Step 3.5 Flash Base at Q4_K_M (130.5 GB), including Mac Pro M2 Ultra (192 GB), Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio M2 Ultra (192 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.