Step 5 Preview — Hardware Requirements & GPU Compatibility
ChatFunctions
Step 5 Preview is StepFun's flagship model for agentic work, covering repository-level software engineering, long-horizon tool use, and professional knowledge work, with a particular focus on finance. It is a sparse mixture-of-experts model with about 600 billion total parameters and 27 billion active per token, and it accepts text and images and produces text. A Q4 build needs roughly 345 GB of memory, so local use means a multi-GPU server or a very large unified-memory machine. The context window is up to 1 million tokens, with output also allowed up to 1 million tokens. It is released under a custom license and was published on September 20, 2026. The card lists FP8 and BF16 checkpoints and serving through vLLM, SGLang, or Transformers. Its FP8 examples use one node of four B300 GPUs or eight H200 GPUs. It recommends temperature 1.0 and top_p 0.95, and its results use the High reasoning setting.
Step 5 Preview runs on a 512 GB+ Mac or a multi-GPU server
Best local pick → Mac Studio (M3 Ultra, 512GB) · Q4_K_M · 407.5 GB · ~20 tok/s
Macs from 256 GB (IQ2_XXS · ~31 tok/s)
Score 128#8 of 101 open#6 of 98 that fit ≤512 GBOther
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
llmrun Composite Score →
128.0
- 90% range 124.3–131.8
- #8 of 101 open models
- #6 of 98 that fit ≤512 GB
Run Step 5 Preview
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve stepfun-ai/Step-5-Preview --max-model-len 16384 --tensor-parallel-size 8Size the nodes from the table above
Download: Step-5-Preview-Q4_K_M.gguf362.60 GBsmalinin
Other ways
each opens step-by-stepRun in cloud
Rent 6× H100 SXM · Q4_K_M
- Estimated hourly price
- ~$16.14/hr
- est., whole server
- GPUs
- 6 GPUs
- 480 GB of VRAM
How Much VRAM Does Step 5 Preview Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | Runs on | |
|---|---|---|---|---|
| 182.8 GB | 201.3 GB | 2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss | Runs on B200 192 GB | |
| 282.5 GB | 301 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Runs on 5× H100 SXM 80 GB | |
| 299.1 GB | 317.6 GB | 3-bit · i-quant medium — Importance-weighted 3-bit, medium | Runs on 5× H100 SXM 80 GB | |
| 324.1 GB | 342.5 GB | 3-bit medium — 3-bit medium quantization | Runs on 5× H100 SXM 80 GB | |
| 398.9 GB | 417.3 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Runs on 6× H100 SXM 80 GB | |
| 473.6 GB | 492.1 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Runs on 7× H100 SXM 80 GB | |
| 548.4 GB | 566.9 GB | 6-bit, very good — 6-bit quantization, very good quality | Runs on 8× H100 SXM 80 GB | |
| 664.8 GB | 683.2 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Runs on 6× H200 SXM 141 GB | |
| 1329.5 GB | 1348 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Runs on 8× B200 192 GB |
Selected Q4_K_M
4-bit medium quantization407.5GBat 16K context
- Weights
- 397.3 GB
- KV cache
- 9.8 GB
- Overhead
- 0.3 GB
Server-class: ~407.5 GB at 16K context · about 6× H100 SXM 80 GB
Which Hardware Can Run Step 5 Preview?
Q4_K_M · 398.9 GBStep 5 Preview (Q4_K_M) requires 398.9 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 519+ GB is recommended. No consumer GPU has enough memory. 1 device with unified memory can run Step 5 Preview, including Mac Studio (M3 Ultra, 512GB).
192+ GB
Q4_K_M · best fit IQ2_XXS (191.4 GB)
6 GPUs · 1 Mac class · 2 device families
B200~151Simulate output on B200B300~151Simulate output on B300Instinct MI350X~50Simulate output on Instinct MI350XM3 Ultra256–512 GB~31Simulate output on Mac Studio (M3 Ultra, 256GB)8x H100 SXM5~181Simulate output on NVIDIA DGX H1008x A100 SXM4~172Simulate 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 192 GB: too small for this model (needs ≥ 191.4 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →
How good is Step 5 Preview?
The llmrun Score merges 3 public benchmark boards into a single figure, where 100 represents the average for open models.
Score vs VRAM in the ≤512 GB tier
- Step 5 Preview
- Other models
- Best Score for its VRAM
#6 of 98 that fit ≤512 GB
Each dot is a model at the best quant that fits ≤512 GB. The line joins the models with the best Score for their VRAM; a dot below it is beaten by a model that needs no more memory.
Full ≤512 GB ranking →Benchmarks
Here are all 3 boards that have results for this model, sorted by their weight in the llmrun Score; badges show who ran the test.
- SciCode
Epoch AI · Sep 2026Lab-reported
Lab-reported58.9%
#4/68 open
- ProofBench
Epoch AI · Sep 2026Lab-reported
Lab-reported42.0%
#9/27 open
- CritPt
Epoch AI · Sep 2026Lab-reported
Lab-reported20.9%
#3/68 open
In the ≤512 GB class
98 models fit ≤512 GBSame Score, only models that fit ≤512 GB at their best quant: this model's peers by memory need.
- #1
DeepSeek V4.1 Flash 1017DeepSeek
133.2
458.4 GB
- #2
MiMo V2.6 Pro RLXiaomi
133.2
499.3 GB
- #3
MiMo V2.6 Flash RLXiaomi
131.9
186.5 GB
- #4
GLM 5.3Z.ai
130.1
453.8 GB
- #5
DeepSeek V4.1 FlashDeepSeek
129.4
458.4 GB
Official models with an llmrun Score, each answering one question; sizes are each model's default quant (usually Q4_K_M) with a 16K-token context.
Best alternative at this size
DeepSeek V4.1 Flash 1017
DeepSeek
- 5.2 higher Score
- 51 GB more VRAM
Score 133.2 · 458.4 GB
Fits in less VRAM
GLM 5.3 Flash
Z.ai
- 3.9 lower Score
- 202.3 GB less VRAM
Score 124.1 · 205.2 GB
Other vendor
GLM 5.3
Z.ai
- 2.1 higher Score
- 46.3 GB more VRAM
Score 130.1 · 453.8 GB
Similar models
10| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| MiMo V2.6 Flash RLXiaomi | 131.9 | 186.5 GB | —needs > 24 GB | similar score |
| DeepSeek V4.1 FlashDeepSeek | 129.4 | 458.4 GB | —needs > 24 GB | similar score |
| DeepSeek V4 Flash 0731DeepSeek | 127.6 | 183 GB | —needs > 24 GB | similar score |
| GLM 5.2Z.ai | 122.9 | 453.8 GB | —needs > 24 GB | similar score |
| Qwen3.8 27BAlibaba | 121.9 | 17.9 GB | ~38 | similar score |
| GLM 5.1Z.ai | 117.8 | 454.1 GB | —needs > 24 GB | similar score |
| Qwen3.5 397B A17BAlibaba | 113.0 | 242.6 GB | —needs > 24 GB | similar score |
| GLM 5Z.ai | 111.8 | 454.1 GB | —needs > 24 GB | same size |
| DeepSeek V3.2DeepSeek | 110.4 | 412.7 GB | —needs > 24 GB | same size |
| NVIDIA Nemotron 3 Ultra 550B A55B BF16NVIDIA | 109.6 | 378.4 GB | —needs > 24 GB | same size |
Specifications
- Publisher
- StepFun
- Parameters
- 604.3B
- Active parameters
- 27B per token
- Architecture
- Step5ForConditionalGeneration
- Context Length
- 1,048,576 tokens
- Vocabulary Size
- 128,896
- Release Date
- 2026-09-20
- License
- Other
Frequently Asked Questions
How much VRAM does Step 5 Preview need?
Step 5 Preview requires 398.9 GB of VRAM at Q4_K_M, or 1329.5 GB at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 604.3B × 4.8 bits ÷ 8 = 362.6 GB
KV Cache + Overhead ≈ 36.3 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 5 Preview?
No — Step 5 Preview requires at least 182.8 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
What's the best quantization for Step 5 Preview?
For Step 5 Preview, Q4_K_M (398.9 GB) offers the best balance of quality and VRAM usage. Q5_K_M (473.6 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 182.8 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Step 5 Preview on a Mac?
Yes — Mac Studio (M3 Ultra, 256GB) and 1 other Mac can run Step 5 Preview, the smallest at IQ2_XXS. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 398.9 GB of usable unified memory (RAM minus macOS overhead), from Mac Studio (M3 Ultra, 512GB).
Can I run Step 5 Preview locally?
Yes — Step 5 Preview can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
What's the download size of Step 5 Preview?
At Q4_K_M, the download is about 362.60 GB. The full-precision BF16 version is 1208.68 GB. The smallest option (IQ2_XXS) is 166.19 GB.
Which GPUs can run Step 5 Preview?
No single consumer GPU has enough VRAM to run Step 5 Preview at Q4_K_M (398.9 GB). Multi-GPU or professional hardware is required.
Which devices can run Step 5 Preview?
1 device with unified memory can run Step 5 Preview at Q4_K_M (398.9 GB), including Mac Studio (M3 Ultra, 512GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.