StepFun·Step

Step 3

StepFun’s multimodal reasoning foundation model, co-designed around MoE and MFA/AFD system components with 321B total and 38B active parameters.

LegacyLanguageReasoningCodingMultimodalOpen weights

MODEL NUMBER393

First public date: 2025-07-31
Organization: StepFun
Family: Step

Specifications

Only published specifications are shown.

Total parameters: 321B↗
Active parameters: 38B↗
Architecture: MoE↗
Context: 65,536 tokens↗
Input modalities: Text · Image↗
Output modalities: Text↗
Open weights: Yes↗
License: Apache-2.0↗
Status: Legacy

Overview

Release dates

July 31, 2025 — Step 3 multimodal reasoning foundation model was open-sourced. ↗

Why it mattered

Step 3 is a key point where StepFun’s public-weight path moved from early product models to efficient multimodal reasoning and model-system co-design.

Specifications alongside predecessor

Compared with Step-2 ↗

PreviousStep-2— · — ctx

CurrentStep 3321B · 66K ctx

Architecture & capabilities

Model capabilities

ReasoningCodingTool use

Release history

2025-07-31Open weights

Step 3 multimodal reasoning foundation model was open-sourced.

View source

Related models

Select a card to view the related model.

PredecessorStep-2View details →SiblingSTEP3-VL-10BView details →SuccessorStep 3.5 FlashView details →

Sources

Step3: Cost-Effective Multimodal IntelligenceStepFun · Published 2025-07-31 · Checked 2026-09-22

Official · Verified 02

Step3 official repositoryStepFun · Checked 2026-09-22

Official · Verified

Suggest a correction

Copy this template, add the field, proposed value, and primary source, and share it with the archive maintainer.

Model ID: step-3
Page: https://llmlineage.com/en/models/step-3
Field:
Proposed value:
Primary source URL:
Reason: