AI Index

Step 3.7 Flash

Language StepFun Partly verified Released May 29, 2026
  • In TextImageVideo Out Text
  • Reasoning· low, medium, high
  • Tool use
  • Open weights

Prices

2 offerings · prices as listed
StepFun API Official Global · USD step-3.7-flash
Conditions/ 1M tokensInputCache readOutput
Standard $0.20$0.040$1.15

Prices from StepFun: pricing details (checked Nov 2, 2026)

StepFun API Official China · CNY step-3.7-flash
Conditions/ 1M tokensInputCache readOutput
Standard ¥1.35¥0.27¥8.10

Prices from 阶跃星辰：定价明细 (checked Nov 2, 2026)

Standard: the price a request with no conditions pays (the headline price elsewhere on the site).

Details

Knowledge cutoff
Mar 1, 2026
Licence
Apache 2.0
Family
step-flash
Model ID

Not stated on StepFun’s pages: max input, reasoning can be off, max thinking budget, structured output, built-in tools, aliases.

Run locally

Weights: stepfun-ai/Step-3.7-Flash
  • Parameters 201.4B
  • Active 3.2B (estimated) · 288 experts, 8 per token
  • Native precision BF16
  • Layers 45
  • KV cache per token 180 KB
  • Config positions 256K (config.json max_position_embeddings)
  • API context 250K (maker’s API)
My hardware GeForce RTX 4090 24 GB · 1008 GB/s · 64 GB DDR5 dual channel · 89.6 GB/s
On your hardware · GeForce RTX 4090
Doesn’t fit IQ4_XS (automatic) 8K context
WeightsKV cacheOverheadUsable
Needs
≈ 104.1 GB (weights 97.8 (maker’s file) + KV cache 1.4 + overhead 4.9)
Usable
21.6 GB (90% of its memory)
Try
IQ3_XXS runs (needs system memory), but below 4 bits quality drops noticeably
PrecisionWeightsNeeds at 8KOn your hardwareLongest context
BF16 (native)maker’s file375.1 GB≈ 395.2Doesn’t fit—
Q8_0maker’s file198.5 GB≈ 209.8Doesn’t fit—
FP8maker’s file197.9 GB≈ 209.2Doesn’t fit—
Q6_K154.0 GB≈ 163.1Doesn’t fit—
Q5_K_M133.4 GB≈ 141.5Doesn’t fit—
Q5_0128.9 GB≈ 136.8Doesn’t fit—
Q4_K_M113.9 GB≈ 121.0Doesn’t fit—
NVFP4maker’s file120.4 GB≈ 127.8Doesn’t fit—
Q4_0105.5 GB≈ 112.2Doesn’t fit—
IQ4_XSmaker’s file97.8 GB≈ 104.1Doesn’t fit—
Q3_K_Mmaker’s filequality drops87.4 GB≈ 93.1Doesn’t fit—
IQ3_XXSmaker’s filequality drops70.6 GB≈ 75.5Needs system memory 47–70 tok/s—
Q2_Kquality drops70.3 GB≈ 75.2Needs system memory 47–71 tok/s—
IQ2_XXSquality drops48.3 GB≈ 52.1Needs system memory 81–121 tok/s—

Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.

Compared with other hardware

Each at the highest precision it runs. Click one for details.

Smallest that runs it (Q4_K_M, 8K): GPU NVIDIA H200 SXM; unified memory Apple M5 Ultra (64-core GPU) · 256 GB.

Run commands

llama.cpp
llama-server -hf stepfun-ai/Step-3.7-Flash-GGUF:IQ4_XS -c 8192
Ollama
ollama run hf.co/stepfun-ai/Step-3.7-Flash-GGUF:IQ4_XS
Memory to fine-tune
Full≈ 3003.0 GBDoesn’t fit your hardware
LoRA≈ 377.9 GBDoesn’t fit your hardware
QLoRA≈ 108.3 GBDoesn’t fit your hardware

Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.

Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated

Cited scores

From third-party leaderboards, not measured by us

This model isn’t on any leaderboard we cite yet.

Changes

All changes →

No changes recorded yet.

Sources

Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data

StepFun · China · stepfun.com

Snapshots, the same family and other models from this provider