ai.canshu.com
中文
VRAM Calculator Advisor & TCO API Pricing Compare Models Leaderboard GPUs

Search models, GPUs, and tools

↑↓ Navigate · Enter Open · ★ Watchlist prioritized ai.canshu.com
StepFun 阶跃星辰
Open-Weight StepFun 阶跃星辰 · MoE 201B / A11B · Text+Image+Video · BF16 / FP8

Step 3.7 Flash (201B MoE) Architecture Parameters, Benchmarks & VRAM

Apache-2.0 multimodal agentic MoE (201B/A11B) scoring 56.3% on SWE-Bench Pro.

🧬 Step 3.7 Flash (201B MoE) Architecture & Tensor Specifications

Verified directly against HuggingFace (stepfun-ai/Step-3.7-Flash) config.json & Model Card

HF Class: Step3p7ForConditionalGeneration

1. Backbone & Dimensions

step3p7
Total / Active Params
201B / 11B
Hidden Size (d_model)
4,096
FFN Intermediate Size
11,264
MoE Expert Dim
1,280
MoE Experts
288 Routed · Top-8
Activation / Norm
SiLU (SwiGLU) · ε=1e-6
Native Tensor Dtype
BF16 / FP8

2. Attention & Layer Layout

HYBRID
Num Hidden Layers
45 Layers (12 Global KV)
Hidden Layer Layout
45-Layer Sparse MoE (288 Experts, Top-8 Active)
Attention Heads (Q / KV)
64 Q / 8 KV (8:1 GQA)
Head Dimension
128

3. Context, RoPE & Tokenizer

Tokenizer
Native Context Length
262,144 tokens
RoPE Configuration
RoPE (原生 262,144 Context)
Vocab Size
128,896 (Untied)

📋 Step 3.7 Flash (201B MoE) Official API & System Card Specifications

Verified against official StepFun 阶跃星辰 developer documentation & pricing

API Model ID: step-3.7-flash

1. Identity & Capabilities

API Spec
API Model ID
step-3.7-flash
Release Date
2026-05-29
Weight License
Open-Weight (API Spec)
Native Reasoning
✅ Supported
Function Calling & JSON Schema
✅ 原生支持
Batch API 异步批处理
✅ Supported (50% off)

2. Context, Output & Modalities

Limits
Max Context Window
256,000 tokens (256K)
Max Output Tokens
256,000 tokens
Input Modalities
Text + Image + Video
Output Modalities
Text
Direct China Access
✅ Direct CN + RMB
Deployment Mode
Open-Weight + Cloud API

3. Official Pricing & Cache

CNY / 1M
Input / 1M Tokens
¥ 1.35
Cached Input / 1M
¥ 0.27 -80%
Output / 1M Tokens
¥ 8
Tiered Pricing Notes
StepFun: Official StepFun CN platform

🎛️ Official Sampling & Reasoning Parameters

Recommended generation settings & API parameter constraints

Official Default Sampling
temperature=0.7 top_p=0.95 max_tokens=32,768

📊 Step 3.7 Flash (201B MoE) Full Benchmark Scorecard

4 public benchmarks · ranks among 154 models · synced 2026-11-20

⚔️ Compare head-to-head

Source: ZeroEval / llm-stats.com (✓ = third-party verified, others vendor-reported).

⚡ Step 3.7 Flash (201B MoE) Interactive VRAM & Token Speed Calculator

Concurrency (Batch):
MoE 201B / A11B·45L (d=4096)·64Q/8KV·Ctx 256K·Text+Image+Video
Runnable Models →
512 GB Unified·1200 GB/s·Safe Usable 460.8 GB
Smooth Fit126.47 / 460.8 GB~111.4 tok/s
3. Weight Quantization
4. Per-Request Context16K (16,384 tok)
Verified: 2026-11-20·Roofline bandwidth/compute estimate (Batch=1)
Smooth Fit
126.47/ 460.8 GB usable

Physical 512 GB · Load 27.4%

Theoretical Decode Speed
~111.4tokens/s

MoE: reads only 6.67 GB active weights/step

Max Context at Batch=1
256K/ 256K

Enable Q8/Q4 KV quantization to scale batch concurrency

Weights: 121.86 GBKV Cache (1×16K): 0.88 GBOverhead: 3.74 GB
Reserves ~10% unified memory for OS
🚀 Blazing Fast (Instant CoT)~4.5s for a 500-token response
Reference Launch CLI:
OLLAMA_CONTEXT_LENGTH=16384 ollama run step-3-7-flash-200b

Mac Studio M5 Ultra 512G Fits Step 3.7 Flash (201B MoE) · Cost & TCO

Approx. $9,499 (~$19/GB VRAM) · 190W TDP

📊 Step 3.7 Flash (201B MoE) Quantization VRAM Breakdown (16K Context)

Quantization Bits/Weight Weights KV Cache (16K) Total VRAM Quality Note
FP16 / BF16 16 bpw 402 GB 0.88 GB 412.22 GB Lossless Full Precision
FP8 / Q8_0 8.5 bpw 213.56 GB 0.88 GB 220.01 GB 99.9% Near-Lossless
GGUF Q6_K 6.56 bpw 164.82 GB 0.88 GB 170.29 GB Ultra High Quality
GGUF Q5_K_M 5.69 bpw 142.96 GB 0.88 GB 148 GB High Balanced Quality
Q4_K_M / AWQ 4.85 bpw 121.86 GB 0.88 GB 126.47 GB Sweet Spot (Recommended)
GGUF IQ4_XS 4.25 bpw 106.78 GB 0.88 GB 111.09 GB Compact 4-bit
GGUF Q3_K_M 3.91 bpw 98.24 GB 0.88 GB 102.38 GB Moderate Perplexity Loss
GGUF Q2_K 3.15 bpw 79.14 GB 0.88 GB 82.9 GB Extreme Compression

🖥️ GPU & Apple Silicon Compatibility for Step 3.7 Flash (201B MoE) (Q4_K_M)

Frequently Asked Questions

What are the exact architecture parameters of Step 3.7 Flash (201B MoE)?

Step 3.7 Flash (201B MoE) uses Step3p7ForConditionalGeneration with 45 layers (45-Layer Sparse MoE (288 Experts, Top-8 Active)), hidden_size=4096, 64 Q heads / 8 KV heads (head_dim=128), and vocab_size=128,896.

How much VRAM does Step 3.7 Flash (201B MoE) require?

At 4-bit quantization (Q4_K_M, 16K context), Step 3.7 Flash (201B MoE) requires 126.47 GB VRAM. Full FP16 requires 412.22 GB.