LLM
GA
Open weights
Step-3
Open-weight multimodal reasoning
Release2025-07
Context66K
Max output—
Parameters321B
Local / VRAM400–640 GB
LicenseOpen weights
Overview
Step-3 is a model in the LLM category from StepFun. It was released in 2025-07. The vendor is based in China. Best for: Open-weight multimodal reasoning.
Strengths
- Apache-2.0
- Vision input
- Strong math
Limits
- 64K context
- Heavy hardware
Specifications
| Vendor | StepFun |
|---|---|
| Category | LLM |
| Release | 2025-07 |
| Status | GA |
| License | Open weights |
| License name | Apache-2.0 |
| Commercial use | Allowed |
| Parameters | 321B |
| Active parameters | 38B |
| Architecture | MoE VLM |
| Context window | 66K |
| Input modalities | Text, Image |
| Output modalities | Text |
| Prompt types | Text |
| Negative prompts | No |
| Max outputs | 1 |
| Local deployment | Yes |
| VRAM | 400–640 GB |
| System RAM | 512–1024 GB |
| API access | Yes |
| Pricing | Free weights |
| Confidence | High |
| Vendor country | China |
| Vendor type | Open lab |
Pricing & hardware
Free weights
This model can be self-hosted or run locally. Approximate GPU memory: 400–640 GB. Approximate system RAM: 512–1024 GB. API access is available.