MODEL LINEAGE
中文
Model index
StepFun·Step

Step 3.5 Flash

A sparse MoE model for fast reasoning, coding, and agent workflows, with 196B total parameters, 11B active per token, and 256K context.

CurrentLanguageReasoningCodingAgentOpen weights
MODEL NUMBER395
First public date
2026-02-02
Organization
StepFun
Family
Step
01

Specifications

Only published specifications are shown.

Total parameters
196B↗
Active parameters
11B↗
Architecture
MoE↗
Context
256,000 tokens↗
Input modalities
Text↗
Output modalities
Text↗
Open weights
Yes↗
License
Apache-2.0↗
Status
Current
02

Overview

Release dates

— Step 3.5 Flash launched with open weights. ↗

Why it mattered

Step 3.5 Flash combines sparse activation, sliding-window attention, and multi-token prediction into an open model for real-world agent tasks.

03

Specifications alongside predecessor

Compared with Step 3 ↗

PreviousStep 3321B · 66K ctx
CurrentStep 3.5 Flash196B · 256K ctx
04

Architecture & capabilities

Attention

Sliding Window AttentionFull Attention

Model capabilities

ReasoningCodingTool useLong context

Unclassified tags

Fast action
05

Release history

Open weights

Step 3.5 Flash launched with open weights.

View source
06

Related models

Select a card to view the related model.

PredecessorStep 3View details →SuccessorStep 3.7 FlashView details →
07

Sources

01
Step 3.5 Flash: Fast Enough to Think. Reliable Enough to Act.StepFun · Checked 2026-09-22
Official · Verified02
Step-DeepResearch official repository release noteStepFun · 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.