// Open-model VRAM table · family
StepFun Step: local deployment & VRAM
12 open-weight models from StepFun (23 official repos incl. quantised builds). At 4-bit they range from 1.7 GB (GOT-OCR-2.0-hf) to 412 GB (Step-5-Preview).
official orgs: stepfun-ailatest 2026-09-20memory = estimate
01Text LLM (2)
| Model | Params | Context | BF16 | 4-bit | Smallest GPU (4-bit) | Mac (4-bit) | Licence | Released |
|---|---|---|---|---|---|---|---|---|
| Step-5-Preview | 600B / 27B | 1M | 1.33 TB | 412 GB | 8 × H100 (640GB) | — | Custom licence – read it | 2026-09-20 |
| Step-3.5-Flash | 196B / 11B | 256K | 436 GB | 135 GB | 4 × RTX PRO 6000 (384GB) | Mac 256GB | Commercial OK | 2026-02-01 |
02Vision-language (5)
| Model | Params | Context | BF16 | 4-bit | Smallest GPU (4-bit) | Mac (4-bit) | Licence | Released |
|---|---|---|---|---|---|---|---|---|
| Step-3.7-Flash | 198B / 11B | 256K | 440 GB | 136 GB | 4 × RTX PRO 6000 (384GB) | Mac 256GB | Commercial OK | 2026-05-23 |
| Step3-VL-10B | 10.2B | 64K | 24.1 GB | 8.5 GB | 12GB GPU | Mac 16GB | Commercial OK | 2026-01-13 |
| step3 | 321B / 38B | — | 740 GB | 247 GB | 4 × RTX PRO 6000 (384GB) | Mac 512GB | Commercial OK | 2025-07-28 |
| GOT-OCR-2.0-hf | 561M | — | 2.5 GB | 1.7 GB | 8GB GPU | Mac 16GB | Commercial OK | 2024-11-22 |
| GOT-OCR2_0 | 716M | 32K | 2.9 GB | 1.8 GB | 8GB GPU | Mac 16GB | Commercial OK | 2024-09-12 |
03Omni / audio LLM (5)
| Model | Params | Context | BF16 | 4-bit | Smallest GPU (4-bit) | Mac (4-bit) | Licence | Released |
|---|---|---|---|---|---|---|---|---|
| Step-Audio-R1.1 ↗ | 33.5B | 64K | 76.3 GB | 25.0 GB | RTX 5090 32GB | Mac 64GB | Commercial OK | 2026-01-14 |
| Step-Audio-EditX | 3.53B | — | 11.5 GB | 6.1 GB | 8GB GPU | Mac 16GB | Licence not stated | 2025-10-29 |
| Step-Audio-2-mini | 8.31B | 16K | 19.3 GB | 6.5 GB | 8GB GPU | Mac 16GB | Commercial OK | 2025-08-28 |
| Step-Audio-AQAA | 136.7B | — | 337 GB | 127 GB | 4 × RTX PRO 6000 (384GB) | Mac 256GB | Commercial OK | 2025-06-11 |
| Step-Audio-Chat | 131.9B | — | 326 GB | 124 GB | H200 141GB | Mac 256GB | Commercial OK | 2025-02-13 |
04Other families
Qwen · IBM Granite · NVIDIA Nemotron · Gemma · Mistral / Magistral / Devstral · GLM / CogVideoX (Z.ai) · DeepSeek · Ai2 OLMo / Molmo · Ant Ling / Ring / Ming · Liquid LFM · MiniCPM / VoxCPM · Tencent Hunyuan / Hy · Phi · Llama · Meituan LongCat · Jina embeddings & rerankers · Wan · Kimi · Cohere Command / Aya · Xiaomi MiMo · FLUX · Microsoft VibeVoice / Fara / Harrier · InternLM / Intern-S · MiniMax · Baidu ERNIE · NVIDIA Parakeet / Canary speech · Stable Diffusion / Stable Audio · Yi (01.AI) · ByteDance Seed · SmolLM / SmolVLM · BGE embeddings & rerankers · LTX Video · FunAudioLLM speech · gpt-oss · poolside Laguna · Whisper · Baichuan · Thinking Machines Inkling · xAI Grok · rednote dots
Totals include 10% + 0.5GB overhead and one 8K-token KV cache for text models. Estimates only. Method →