Step 3.7 Flash 198B-A11B
StepFun's local-inference-focused multimodal MoE has a 198B language backbone with about 11B active parameters; the full official archive is about 201B including vision components. It supports 256K context and officially targets high-memory unified systems such as Mac Studio and DGX Station.
StepFunMixture of expertsagenticreasoning9 machines can run it
Total parameters
201.365B
Official spec sets memory need
Active parameters
11B
Official spec sets decode speed
Context
256K
Official spec tokens
4-bit weights
110 GB
Calculated before KV cache
This is a sparse mixture of experts. All 201.365B of weights must be available to the runtime, but only 11B are read per generated token. Ordinary runtimes keep the full quantized model in memory; a runtime-specific SSD-streaming artifact can retain a smaller working set and fetch expert data on demand, trading speed for capacity.
Model specification
Publisher
StepFun Official spec
Architecture
Sparse mixture of experts Official spec
Total parameters
201.365B Official spec
Active parameters
11B per token Official spec
Context length
262,144 tokens Official spec
Attention ⓘ
MHA · 64 KV heads × 128 dim × 45 layers Official spec
Licence
Apache 2.0 Official spec source
Specification confidence ⓘ
High · verified 2026-09-21 Official spec source
Released
23 May 2026 Official spec
Official source
Quantizations and memory
| Quantization | Format | Bits/weight | Resident | Download | Quality kept |
|---|---|---|---|---|---|
MLX 4-bit | mlx | 4.5 | 109.7 GB | 109.7 GB | 98.0% |
Q4_K_Mdefault | gguf | 4.85 | 117.1 GB | 117.1 GB | 98.5% |
Q5_K_M | gguf | 5.69 | 137.4 GB | 137.4 GB | 99.3% |
Q8_0 | gguf | 8.5 | 203.2 GB | 203.2 GB | 99.9% |
BF16 | safetensors | 16 | 375.1 GB | 375.1 GB | 100.0% |
Generic weight sizes are computed from the parameter count and bits per weight plus a format-specific overhead. Runtime-specific artifacts use their published resident and download footprints; streamed models can therefore require much more disk than memory. Quality retention is an assumption, not a measured evaluation.
Coding & quality benchmarksCompare coding results →
No published benchmark results have been imported for this model yet. This is missing evidence, not a score of zero.
We show source metrics rather than deriving one opaque quality number. Different benchmarks measure genuinely different things, and collapsing them into a single score would hide exactly the disagreements worth seeing.
Recommendations
Cheapest that can run it
~4.7–6.7 t/s · $1,531
Lowest purchase price among configurations where the model fits at some quantization in our catalogue. Speed is not considered.
Cheapest above 20 t/s
~48.6–69.9 t/s · $5,674
20 tokens/second is roughly the point at which generation keeps pace with reading. Below it, interactive use feels like waiting.
Cheapest above 40 t/s
~48.6–69.9 t/s · $5,674
40 tokens/second is the threshold most people describe as comfortable for coding agents, where output arrives faster than you can review it.
Fastest with real measurements
Nothing in the database qualifies.
Highest throughput among configurations with an actual published measurement rather than our estimate.
Best throughput per purchase-price unit
~84.1–121 t/s · $8,647
Highest decode tokens/second per 1,000 units of the displayed purchase currency. Ignores running costs and resale — see the economics section for the full picture.
Hardware that runs Step 3.7 Flash 198B-A11B0 measured, 9 estimated
9 of 9 rows
| Hardware↕ | Quant | Memory↕ | Decode▼ | Prefill↕ | Context | Price↕ | Fit | Confidence↕ |
|---|---|---|---|---|---|---|---|---|
| NVIDIA DGX H200 (8x H200, 1,128 GB) NVIDIA · 1,128 GB · 38,400 GB/s | Q8_0recommended | 224.4 GB | ~431–620 t/s | ~56850–118070 t/s | 256K | $359,430 | Comfortable | Estimated |
| Dell Precision 7960 Rack (2x RTX PRO 6000, 192 GB) Dell · 192 GB · 3,584 GB/s | Q4_K_Mrecommended | 133.3 GB | ~143–205 t/s | ~1360–2830 t/s | 16K | $36,663 | Comfortable | Estimated |
| Lenovo ThinkStation PX (4x RTX PRO 6000, 384 GB) Lenovo · 384 GB · 7,168 GB/s | Q8_0recommended | 222.8 GB | ~126–181 t/s | ~2730–5670 t/s | 64K | $58,553 | Comfortable | Estimated |
| Mac Studio M5 Ultra 256 GB Apple · 256 GB · 1,200 GB/s | Q5_K_Mrecommended | 153.8 GB | ~84.1–121 t/s | ~617–1280 t/s | 32K | $8,647 | Comfortable | Estimated |
| NVIDIA DGX Station GB300 (748 GB) NVIDIA · 748 GB · 7,100 GB/s | Q8_0recommended | 221.6 GB | ~58.9–84.8 t/s | ~7600–15780 t/s | 256K | $90,082 | Comfortable | Estimated |
| Mac Studio M5 Ultra 512 GB Apple · 512 GB · 1,200 GB/s | Q8_0recommended | 221.6 GB | ~58.3–83.8 t/s | ~617–1280 t/s | 128K | $11,350 | Comfortable | Estimated |
| Mac Studio M3 Ultra 256 GB Apple · 256 GB · 819 GB/s | Q5_K_Mrecommended | 153.8 GB | ~48.6–69.9 t/s | ~149–309 t/s | 32K | $5,674 | Comfortable | Estimated |
| Mac Studio M3 Ultra 512 GB Apple · 512 GB · 819 GB/s | Q8_0recommended | 221.6 GB | ~33.2–47.8 t/s | ~149–309 t/s | 128K | $8,287 | Comfortable | Estimated |
| CPU-only workstation (Ryzen 9950X, 192 GB DDR5) Generic · 192 GB · 90 GB/s | Q4_K_Mrecommended | 132.9 GB | ~4.7–6.7 t/s | ~8.8–18.4 t/s | 32K | $1,531 | Comfortable | Estimated |
Submit a benchmarkContributions are reviewed before publication.
Similar models
Qwen3.8 Flash Next
Alibaba Qwen · 180B (6B active) · Qwen Community 1.0
MiMo-V2.6 Flash 309B-A15B
XiaomiMiMo · 309B (15B active) · MIT
Nemotron 3 Super 120B-A12B
NVIDIA · 123.611B (12B active) · NVIDIA Nemotron Open Model License
MiniMax M3
MiniMax · 427.04B (23B active) · Apache 2.0
DeepSeek-V4.1 Flash
DeepSeek · 763.205B (16B active) · MIT
MiMo-V2.6 Pro 1.02T-A42B
XiaomiMiMo · 1.0T (42B active) · MIT