Robots Atlas>ROBOTS ATLAS
Step-2
AI Models›Step

Step-2

Step-2 (Step-2-16K) · Family: Step
StepFun’s trillion-parameter MoE language model (2024); 16K context window. Reportedly the first trillion-parameter MoE model from a Chinese company; ~5th on LiveBench.
✓ Active✓ Public accessLLM📁 Step
Context window
16K
tokens
Parameters
~1 bln (bilionowy model MoE)
parameters
Release date
1 July 2024
Access:APIDeployment:☁ Cloud

Overview

Step-2 is a large language model from the Chinese company StepFun (阶跃星辰), released in July 2024. It is a Mixture-of-Experts (MoE) model at trillion-parameter scale — reportedly the first trillion-parameter MoE model developed by a Chinese company.

The model is focused on text processing and supports a 16,000-token context window. According to the LiveBench ranking, Step-2 placed around fifth globally, ahead of GPT-4o and behind o1-mini, among others.

Step-2 is part of the broader StepFun model family, which also includes multimodal models (Step-1.5V), image generators (Step-1X) and newer language models (Step-3, Step-3.5 Flash). The model is offered commercially through StepFun’s platform/API.

StepFun (Shanghai Jieyue Xingchen) was founded in April 2023 by former Microsoft employees (including Jiang Daxin) and is one of China’s “AI Tigers”; its investors include Tencent and Qiming Venture Partners.

Classification
LLM
Family: Step
Applications
ChatbotsCodingQ&A / Question answeringContent generation
Access & deployment
API
Cloud
Weights: Closed
Key parameters
📏 Context: 16K
🧩 Parameters: ~1 bln (bilionowy model MoE)
📥 Input: text

Technical specification

Context window
16K
tokens
Parameters
~1 bln (bilionowy model MoE)
parameters
License
Proprietary
Modalities
⬇ Input
text
⬆ Output
textcode

Capabilities and applications

Native model capabilities
Language modeling
Ability to predict subsequent tokens and generate coherent natural-language text based on the preceding context.
Category: language
Reasoning
The model's ability to reason logically and solve complex problems.
Category: reasoning
Multi-step reasoning
Carrying out multi-step chains of reasoning across long, complex tasks.
Category: reasoning
Coding
Generating, analysing and modifying code in many programming languages. Covers writing functions, debugging, refactoring, code review, and creating tests. Measured by benchmarks such as HumanEval and SWE-bench.
Category: coding
Multilingual
Competence in many natural languages (from a few to over a hundred): understanding, generation, translation, and code-switching within a single conversation. Frontier models support a wide range of languages with comparable quality.
Category: language
Instruction following
Precisely following instructions contained in the prompt: response format, length, style, constraints (e.g. 'reply in six words'). GPT-5.1 significantly improved this capability compared to GPT-5.
Category: language
Application domains
ChatbotsCodingQ&A / Question answeringContent generation

Technical architecture

Core Architecture
TRTransformerMOMoE
Model Form
LLLLM
Training Techniques
PRPretrainingRLRLHFITInstruction Tuning

Sources and related pages