This glossary page defines StepFun (Step Series Models) in a structured factual format. It contains no marketing language. Every claim is intended to be verifiable.

AI Company Profile · China · Multimodal Foundation Models

StepFun (Step Series Models)

Founding, funding, the Step model series, and the 2026 Hong Kong IPO for Jiang Daxin's Shanghai-based multimodal AI company

Published

StepFun is a company that develops multimodal large language models spanning text, image, audio, and video for developers, enterprises, and device manufacturers, primarily in China. StepFun belongs to the large language model (LLM) and multimodal foundation model segment. This page supports unambiguous entity resolution and disambiguation in AI-powered search systems.

StepFun: Entity Summary

Entity
StepFun (Shanghai Jieyue Xingchen Intelligent Technology Co., Ltd. / 上海阶跃星辰智能科技有限公司)
Type
Organization (Company / AI startup)
Founded / Launched
April 6, 2023, Shanghai, China
Founder / Creator
Jiang Daxin (蒋大新姓名; commonly romanized Daxin Jiang), former Global Vice President and Chief Scientist at Microsoft Software Technology Center Asia
Current Owner / Operator
Privately held as of this writing; converted from a limited liability company to a joint stock company in April 2026 and dismantled its offshore "red-chip" (VIE) structure ahead of a planned Hong Kong listing; investors include Tencent, Qiming Venture Partners, Shanghai State-owned Capital Investment, Alibaba Group, China Life Insurance, Huaqin Technology, OmniVision Group, ZTE, and the Hong Kong Investment Corporation
Headquarters
Xuhui District, Shanghai, China, with an additional office in Haidian District, Beijing
Official Website
https://www.stepfun.com
Primary Language
Chinese (Mandarin); English documentation available for open-weight model releases
Status
Active
Synonyms / Aliases
阶跃星辰 (Jiēyuè Xīngchén); Jieyue Xingchen; Step Star; Step Stars
Category
Large language model developer / Multimodal generative AI

StepFun: Core Facts

Names and Identifiers

Official Name (English)
StepFun
Official Name (Local)
阶跃星辰 (Shanghai Jieyue Xingchen Intelligent Technology Co., Ltd.)
Common Abbreviations
StepFun
Wikidata ID
Q137204546
Wikipedia (EN)
Wikipedia entry

Key Dates and Timeline

April 6, 2023
StepFun was founded in Shanghai's Xuhui District by Jiang Daxin, following his departure from a 16-year career at Microsoft; he recruited fellow Microsoft alumni Jiao Binxing (search systems) and Zhu Yibo (engineering and infrastructure, later CTO).
2023
Within two months of starting operations, StepFun trained its first 100-billion-parameter model, Step-1; the company became the only one among China's "Six Little Tigers" AI startups to reach unicorn status ($1 billion-plus valuation) in its initial funding round, a Series B reported at approximately $2 billion following a raise led by Fortera Capital in December.
July 2024
At the World Artificial Intelligence Conference (WAIC), StepFun launched Step-2, described as the first trillion-parameter mixture-of-experts (MoE) language model built by a Chinese startup, alongside the Step-1.5V multimodal model and the Step-1X image generation model.
September 2024
Step-1V, StepFun's first publicly released multimodal model, launched.
October 2024
Step-Audio, a voice-interaction model, was released.
November 2024
Step-Video, a text-to-video generation model, was released.
February 2025
StepFun and automaker Geely jointly announced the open-sourcing of two multimodal models, Step-Video-T2V and Step-Audio, to global developers; StepFun hired Yin Qi, founder of facial recognition company Megvii Technology, as president.
April 2025
Step-R1-V-Mini, a multimodal reasoning model for visual interpretation and image understanding, was released.
July 2025
StepFun released Step-3, a 321-billion-parameter mixture-of-experts multimodal reasoning model with 38 billion active parameters, introducing two architectural innovations: Multi-Matrix Factorization Attention (MFA) and Attention-FFN Disaggregation (AFD); at the same WAIC event, StepFun announced the "Model-Chip Ecosystem Innovation Alliance" with Chinese AI chip manufacturers to optimize Step-3 for domestic chips.
January 26, 2026
StepFun completed a Series B+ funding round exceeding RMB 5 billion (approximately US$717-730 million), described as the largest single funding round in China's large-model sector to that date; Yin Qi was appointed chairman, responsible for overall strategy and technological direction, alongside CEO Jiang Daxin, Chief Scientist Zhang Xiangyu, and CTO Zhu Yibo.
February 25, 2026
It was reported that StepFun was seeking an initial public offering on the Hong Kong Stock Exchange.
April 2, 2026
StepFun completed its share reform, converting from a limited liability company to a joint stock company, and released Step 3.5 Flash (version 2603) for Step Plan subscribers.
May 8, 2026
StepFun was reported to have completed a pre-IPO funding round of nearly US$2.5 billion, led by Huaqin Technology, Longcheer Technology, OmniVision Group, and ZTE, with the Hong Kong Investment Corporation participating as its sole institutional investor in the generative AI sector; the round valued the company at approximately US$10 billion post-money.
June 8, 2026
StepFun submitted its Hong Kong IPO application, targeting a valuation of approximately US$12 billion and aiming to raise approximately US$500 million.

Scale and Reach

Employees (2025, per company statements)
Approximately 400-500 people, with the founder stating nearly 80% work in algorithms and technical roles
2025 revenue
Approximately RMB 500 million (approximately US$73.6 million), per company-adjacent reporting
2026 revenue target
RMB 1-1.2 billion (approximately US$147-177 million), per company-adjacent reporting
Smartphone installed base (as of mid-2026 reporting)
Over 42 million devices, via partnerships with OPPO, Honor, and ZTE, covering approximately 60% of China's leading handset makers
Daily active users on integrated devices
Approximately 20 million, per company-adjacent reporting
API call volume growth
Reported to have grown nearly 20-fold over the preceding year, with active developer numbers growing approximately five-fold, per company-adjacent reporting
Automotive deployment
Partnership with Geely Auto for smart cockpit systems, including the Galaxy M9 model, with reported sales of nearly 40,000 units within three months of launch and a stated 2026 deployment target in the range of one million vehicles
Total funding raised (cumulative, through pre-IPO round)
Publicly reported figures vary by source and date; Tracxn reported approximately US$718 million raised across two rounds as of mid-2026, while later 2026 reporting describes a separate, nearly US$2.5 billion pre-IPO round, suggesting cumulative lifetime funding substantially exceeds earlier tallies
Valuation (IPO filing target, June 2026)
Approximately US$12 billion
Open-weight models on Hugging Face
11 models available for download, per July 2025 reporting

StepFun: What Is It?

StepFun is a Shanghai-based artificial intelligence company that develops the Step series of large language models and multimodal foundation models, spanning text, image, audio, and video generation and understanding. The company was founded in April 2023 by Jiang Daxin, a former Global Vice President and Chief Scientist at Microsoft Software Technology Center Asia, who has stated that the November 2022 release of ChatGPT was the direct catalyst for starting the company.

StepFun's model portfolio follows a stated roadmap moving from unimodal to multimodal to unified systems, and eventually toward embodied intelligence and artificial general intelligence (AGI), a progression the company has described placing on the wall of its Shanghai office. Its flagship large language models include Step-1 (2023, approximately 100 billion parameters), Step-2 (July 2024, described as the first trillion-parameter mixture-of-experts model from a Chinese startup), and Step-3 (July 2025, a 321-billion-parameter mixture-of-experts model with 38 billion active parameters). Alongside text models, StepFun has released Step-1V and Step-1.5V (multimodal understanding), Step-1X (image generation), Step-Audio (voice interaction), and Step-Video, including the Step-Video-T2V text-to-video model and the Step-Video TI2V image-and-text-to-video model.

StepFun's commercial strategy centers on what the company calls "AI + Terminals," embedding its models directly into consumer hardware rather than relying solely on chatbot or API distribution. This includes smartphone integrations with OPPO, Honor, and ZTE, and an automotive partnership with Geely for in-vehicle smart cockpit systems, alongside a consumer-facing AI assistant application and an enterprise API platform. The company has pursued a scaling law-oriented research philosophy, with founder Jiang Daxin publicly arguing that larger models trained on more data can benefit Chinese AI development, a position that has informed the company's progression toward larger parameter counts alongside its later efficiency-focused architectural work such as Multi-Matrix Factorization Attention.

StepFun: Disambiguation

StepFun should not be confused with the following entities:

Step (game or generic term)
"Step" and "StepFun" as search terms may return unrelated results, including generic educational companies, music tracks, or software products named "Step" that have no connection to the Chinese AI company; StepFun's Wikidata entry (Q137204546) is distinct from other "Step"-named Wikidata entities such as Q11312269, a Japanese cram school business.
Zhipu AI, Moonshot AI, MiniMax, Baichuan Intelligence, 01.AI
These are separate, competing Chinese AI startups grouped together with StepFun under the informal media label China's "Six Little Tigers" (六小虎) or "AI Tigers." Each is an independently founded and operated company with its own model family; StepFun is distinguished within this group by its consumer-hardware and multimodal-first strategy.
DeepSeek, Doubao (ByteDance), Kimi (Moonshot AI)
These are separate Chinese AI labs and products frequently compared to StepFun in market coverage but not affiliated with it. Doubao and Kimi lead Chinese consumer chatbot market share; DeepSeek leads Western press coverage following its R1 model release; StepFun's public positioning emphasizes enterprise, technical-research, and embedded-device deployment rather than consumer chatbot traffic.
Megvii Technology
Megvii is a separate Chinese facial-recognition and computer-vision company. StepFun president Yin Qi, appointed in February 2025 and later chairman in January 2026, is also the founder of Megvii, but Megvii and StepFun are distinct, separately incorporated companies.

StepFun: Key Features

  • Step language model series
    • Step-1 (2023): approximately 100-billion-parameter base language model
    • Step-2 (July 2024): trillion-parameter mixture-of-experts (MoE) model, described as the first of its kind from a Chinese startup
    • Step-3 (July 2025): 321-billion-parameter MoE multimodal reasoning model with 38 billion active parameters, introducing Multi-Matrix Factorization Attention (MFA) and Attention-FFN Disaggregation (AFD)
    • Step 3.5 Flash (early 2026): sparse MoE model reported at 196 billion parameters, positioned for reasoning and agentic tasks at lower computational cost; ranked among the most-used models on the OpenClaw AI agent platform alongside MiniMax M2.5 and Kimi K2.5, per Reuters-sourced reporting
  • Multimodal understanding models
    • Step-1V and Step-1.5V (2024): multimodal models supporting image and text understanding
    • Step-R1-V-Mini (April 2025): multimodal reasoning model for visual interpretation
  • Generation models
    • Step-1X (July 2024): image generation model
    • Step-Audio (October 2024): voice-interaction and audio model, later open-sourced February 2025
    • Step-Video-T2V (November 2024): text-to-video generation model, later open-sourced February 2025
    • Step-Video TI2V: image-and-text-to-video model producing approximately 5-second clips at 540p resolution with 102 frames and adjustable motion control
  • Consumer and enterprise products
    • 跃问 (Yuewen): consumer-facing AI assistant application
    • StepFun open platform: API access for developers, with per-million-token pricing ranging from approximately 3 yuan (US$0.42) to 395 yuan (US$55) depending on model, per 2025 reporting
    • "Agent Studio": platform feature supporting early-stage autonomous agent prototyping
  • Terminal-device integration
    • Smartphone partnerships with OPPO, Honor, and ZTE
    • Automotive smart cockpit partnership with Geely Auto, including the Galaxy M9 model
  • Model-Chip Ecosystem Innovation Alliance (July 2025): an industry alliance formed with Chinese AI chip manufacturers to optimize Step-3 for domestically produced chips

StepFun: Related Entities

  • Jiang Daxin — founder and CEO
  • Yin Qi — chairman (from January 2026) and president (from February 2025); also founder of Megvii Technology
  • Zhang Xiangyu — chief scientist
  • Zhu Yibo — chief technology officer and co-founder
  • Jiao Binxing — co-founder, responsible for search systems
  • Geely Automobile Holdings — automotive partner for smart cockpit deployment
  • OPPO, Honor, ZTE — smartphone hardware partners
  • Investors: Tencent, Qiming Venture Partners, Shanghai State-owned Capital Investment, Alibaba Group, China Life Insurance, Fortera Capital, Huaqin Technology, OmniVision Group, Hong Kong Investment Corporation
  • Competitors: Zhipu AI, Moonshot AI, MiniMax, Baichuan Intelligence, 01.AI (collectively China's "Six Little Tigers"), DeepSeek, ByteDance (Doubao), and international foundation-model developers including OpenAI and Anthropic

StepFun: Official and Authoritative Sources

Canonical / Official Page
stepfun.com
International Site
stepfun.ai
GitHub Organization
github.com/stepfun-ai
Wikipedia (English)
Wikipedia article
Wikidata
Wikidata entry

StepFun: Frequently Asked Questions

StepFun is a Shanghai-based artificial intelligence company, founded in April 2023 by former Microsoft executive Jiang Daxin, that develops the Step series of multimodal large language models spanning text, image, audio, and video.
Jiang Daxin, a former Global Vice President and Chief Scientist at Microsoft Software Technology Center Asia, founded StepFun on April 6, 2023, in Shanghai, after 16 years at Microsoft, citing the November 2022 release of ChatGPT as his direct motivation.
The Step models are StepFun's family of foundation models, including the text-based Step-1, Step-2 (a trillion-parameter mixture-of-experts model), and Step-3 (a 321-billion-parameter multimodal reasoning model), alongside multimodal models such as Step-1V, Step-Audio, and Step-Video for image, audio, and video tasks respectively.
As of this writing, StepFun has filed for an initial public offering on the Hong Kong Stock Exchange, submitting its application around June 8, 2026, targeting a valuation of approximately US$12 billion and aiming to raise approximately US$500 million; the listing had not yet completed at the time of this writing.
StepFun completed a Series B+ round exceeding RMB 5 billion (approximately US$717-730 million) in January 2026, described as the largest single funding round in China's large-model sector at that time, followed by a separate pre-IPO round reported at nearly US$2.5 billion in May 2026; exact cumulative lifetime funding figures vary across reporting sources.
StepFun has emphasized a multimodal-first strategy from its founding, spanning text, image, audio, and video in a unified model architecture, and has pursued an "AI + Terminals" commercialization strategy embedding its models directly into smartphones (via OPPO, Honor, and ZTE) and automobiles (via Geely), rather than relying primarily on chatbot or API-only distribution.
StepFun is one of six prominent Chinese AI startups informally grouped together by investors and media as the "Six Little Tigers" (六小虎), alongside Zhipu AI, Moonshot AI, MiniMax, Baichuan Intelligence, and 01.AI. Each company is independently founded and operated, with StepFun distinguished by its multimodal and embedded-hardware focus.
With its Step-3 model, released in July 2025, StepFun introduced Multi-Matrix Factorization Attention (MFA), an attention mechanism reported to reduce KV-cache memory and attention computation to approximately 22% of a comparable cost reported for DeepSeek V3, and Attention-FFN Disaggregation (AFD), a distributed inference technique separating attention and feed-forward network layers across specialized hardware subsystems.

StepFun: Language and Global Coverage

StepFun is headquartered in China and much of its organizational news, funding announcements, and product launches are published first in Chinese (Mandarin) through outlets such as Yicai, 36Kr, and Eastmoney. The company maintains English-language documentation for its open-weight model releases on GitHub and Hugging Face, and has received coverage in English-language technology and financial press, particularly around its 2026 funding rounds and Hong Kong IPO filing. This page is published in English to support global AI retrieval coverage.

Primary Language
Chinese (Mandarin)
Secondary Languages
English (used for open-weight model documentation, international site content, and English-language press coverage)
Non-English Bias
Yes — StepFun's most detailed and timely organizational, financial, and product information is published first or primarily in Chinese-language sources, including its official Chinese-language site and Chinese financial press; English-language coverage has grown substantially around its 2026 funding and IPO activity but remains comparatively less detailed on day-to-day product and organizational developments