Step 3.7 Flash
Step 3.7 Flash is a sparse mixture-of-experts model with image and video understanding, intended for production agents that use tools to investigate information and write code. Its applications include financial document analysis and checking information across sources. Summary written by AI Index from StepFun’s own pages.
- In Input modalities: TextImageVideo Out Output modalities: Text
- Reasoning· low, medium, high
- Tool use
- Open weights
Prices
2 offerings · prices as listedstep-3.7-flash| Conditions/ 1M tokens | Input | Cache read | Output |
|---|---|---|---|
| Standard | $0.20 | $0.040 | $1.15 |
Prices from StepFun: pricing details (opens in a new tab) (checked Nov 9, 2026)
step-3.7-flash| Conditions/ 1M tokens | Input | Cache read | Output |
|---|---|---|---|
| Standard | ¥1.35 | ¥0.27 | ¥8.10 |
Prices from 阶跃星辰:定价明细 (opens in a new tab) (checked Nov 9, 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)
- 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
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 375.1 GB | ≈ 395.2 | ✕Doesn’t fit | — |
| Q8_0maker’s file | 198.5 GB | ≈ 209.8 | ✕Doesn’t fit | — |
| FP8maker’s file | 197.9 GB | ≈ 209.2 | ✕Doesn’t fit | — |
| Q6_K | 154.0 GB | ≈ 163.1 | ✕Doesn’t fit | — |
| Q5_K_M | 133.4 GB | ≈ 141.5 | ✕Doesn’t fit | — |
| Q5_0 | 128.9 GB | ≈ 136.8 | ✕Doesn’t fit | — |
| Q4_K_M | 113.9 GB | ≈ 121.0 | ✕Doesn’t fit | — |
| NVFP4maker’s file | 120.4 GB | ≈ 127.8 | ✕Doesn’t fit | — |
| Q4_0 | 105.5 GB | ≈ 112.2 | ✕Doesn’t fit | — |
| IQ4_XSmaker’s file | 97.8 GB | ≈ 104.1 | ✕Doesn’t fit | — |
| Q3_K_Mmaker’s filequality drops | 87.4 GB | ≈ 93.1 | ✕Doesn’t fit | — |
| IQ3_XXSmaker’s filequality drops | 70.6 GB | ≈ 75.5 | ○Needs system memory 47–70 tok/s | — |
| Q2_Kquality drops | 70.3 GB | ≈ 75.2 | ○Needs system memory 47–71 tok/s | — |
| IQ2_XXSquality drops | 48.3 GB | ≈ 52.1 | ○Needs 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-server -hf stepfun-ai/Step-3.7-Flash-GGUF:IQ4_XS -c 8192
ollama run hf.co/stepfun-ai/Step-3.7-Flash-GGUF:IQ4_XS
Memory to fine-tune
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 2026-11-21. How it’s estimated
Cited scores
From third-party leaderboards, not measured by usThis model isn’t on any leaderboard we cite yet.
Changes
All changes →No changes recorded yet.
Sources
- StepFun: pricing details (opens in a new tab) platform.stepfun.ai · checked Nov 9, 2026
- 阶跃星辰:定价明细in Chinese (opens in a new tab) platform.stepfun.com · checked Nov 9, 2026
- StepFun: Step 3.7 Flash (opens in a new tab) platform.stepfun.ai · checked Nov 9, 2026
- stepfun-ai/Step-3.7-Flash model card (opens in a new tab) huggingface.co · checked Nov 8, 2026
- Step 3.7 Flash — A high-efficiency Flash model for Real-World (opens in a new tab) static.stepfun.com · checked Nov 8, 2026
- models.dev: stepfun/step-3.7-flash (community data, MIT) (opens in a new tab) github.com · checked Sep 30, 2026
- stepfun-ai/Step-3.7-Flash model card (opens in a new tab) huggingface.co · checked Nov 8, 2026
- stepfun-ai/Step-3.7-Flash model card (opens in a new tab) huggingface.co · checked Nov 6, 2026
- stepfun-ai/Step-3.7-Flash config.json (opens in a new tab) huggingface.co · checked Nov 6, 2026
- stepfun-ai/Step-3.7-Flash files (opens in a new tab) huggingface.co · checked Nov 7, 2026
- stepfun-ai/Step-3.7-Flash-FP8 files (opens in a new tab) huggingface.co · checked Nov 7, 2026
- stepfun-ai/Step-3.7-Flash-GGUF files (opens in a new tab) huggingface.co · checked Nov 7, 2026
- stepfun-ai/Step-3.7-Flash-NVFP4 files (opens in a new tab) huggingface.co · checked Nov 9, 2026
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
Related models
Snapshots, the same family and other models from this provider
- Step 3.5 Flash
- Step 3.5 Flash 2603
- Step Audio 2
- Step 1o Audio
- Step 1o Turbo Vision
- Step 5 Preview
- Step Audio R1.5
- StepAudio 2.5 Chat
- Mercury 3 vs Step 3.7 Flash
- GPT-5 Nano vs Step 3.7 Flash
- Ling-3.0 Flash vs Step 3.7 Flash
- Mercury 2.5 vs Step 3.7 Flash
- MiMo-V2.6 Flash vs Step 3.7 Flash
- Seed 2.0 Mini vs Step 3.7 Flash