StepFun
StepFun is a Shanghai-based AI foundation model company that develops and commercializes multimodal large language models (text, vision, speech) via API, open-source releases, consumer apps, and enterprise licensing deals with automotive, cloud, payments, and robotics partners.
- Company typePrivate
- Founded2023
- HeadquartersXuhui, China
- Headcount251–500
- GTM typeB2B and B2C
- OfferingSoftware
What StepFun does
StepFun (Shanghai StepFun Intelligent Technology Co., Ltd., 阶跃星辰) is a Shanghai-based AI foundation model company founded in 2023 that develops and commercializes its own multimodal large models. Its core product line — the Step series — spans language models (Step 1, Step 2, Step 3.5 Flash), the flagship 198B-parameter sparse Mixture-of-Experts vision-language model Step 3.7 Flash with native multimodal input and 256k context, and StepAudio 2.5 Realtime, an end-to-end real-time speech LLM supporting Chinese and English. Models are exposed through three surface areas: a developer-facing Step Open Platform API (Step API) with subscription and usage-based pricing under the Step Plan tier, the Studio application-building platform, and a freemium consumer assistant (阶跃AI) across Mac, Windows, iOS, Android, and web.
The company operates a multi-channel go-to-market combining API-first self-serve developer monetization, freemium consumer reach (Step 1 has been verified across tens of millions of users), and enterprise licensing/embedding partnerships. Named enterprise partners include Geely Auto Group and Ecarx for in-vehicle AI, Huawei Cloud as an AI Model Partner, Alipay for tokenized agentic payments, and Dexmal for embodied robotics. Step 3.7 Flash is released under Apache 2.0 with checkpoints on Hugging Face, OpenRouter, and NVIDIA NIM, combining open-weight distribution with managed API revenue. StepFun is one of China's 'Six Little Tigers' foundation-model startups and has raised over $3.2 billion cumulatively, most recently filing for a Hong Kong IPO at a proposed valuation of up to $12 billion under the chairmanship of Yin Qi, co-founder of Megvii.
StepFun firmographics
Firmographics- Name
- StepFun
- Legal name
- 上海阶跃星辰智能科技股份有限公司 (Shanghai StepFun Intelligent Technology Co., Ltd.)
- Website
- https://stepfun.com
- Company type
- Private
- Founded year
- 2023
- Operating status
- Operating
- Headcount range
- 251–500 employees
- Short description
- StepFun is a Shanghai-based AI foundation model company that develops and commercializes multimodal large language models (text, vision, speech) via API, open-source releases, consumer apps, and enterprise licensing deals with automotive, cloud, payments, and robotics partners.
- Ownership category
- akta.pro rank
StepFun industry classification
Industry- Product category
- Foundation Models / AI Infrastructure
- NAICS
- Computer Systems Design and Related Services (54151), Computer Systems Design and Related Services (5415)
- SIC
- Services-Prepackaged Software (7372), Services-Computer Programming Services (7371)
- akta.pro primary industry
- Foundation Model Developers (LLM/Multimodal Model Labs) (HDAAACAA)
- akta.pro secondary industries
- Multimodal Generation (Text-Image-Video-Audio) & Creative Toolchains (HDAAACAG), Enterprise Foundation Model Integration & APIs (Connectors, Governance, Deployment) (HDAAACAO)
Keywords
Where StepFun is headquartered
LocationHeadquarters
- HQ city
- Xuhui
- HQ country
- China
- HQ region
- Asia
Offices1 record
Markets served
StepFun business model
Business model- GTM type
- B2B and B2C
- Offering type
- Software
- Cost components
- Technology or R&D, Personnel, Infrastructure, Operations, Marketing or Sales
Revenue model
- Step API Platform: Usage-based and subscription-based API access to Step series models (Step 3.7 Flash, Step 3.5 Flash, Step 2, Step 1, Step TTS) through the 阶跃·开放平台. Developers pay for API calls and token usage with the Step Plan subscription option available.
- Consumer AI Applications: Direct-to-consumer 阶跃AI product available across Mac, Windows, iOS, Android, and web platforms, monetized through freemium model with premium features.
- Enterprise Licensing & Integration: Enterprise partnerships embedding StepFun models into automotive (Geely), robotics (Dexmal), cloud platforms (Huawei Cloud), and payment systems (Alipay), with licensing and integration fees.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Subscription | Monthly | Step Plan subscription for API access |
| Usage-based | Pay-as-you-go | Usage-based API pricing |
| Freemium | Monthly | Freemium consumer app |
Go-to-market motion3 records
Distribution channels4 records
Marketing channels4 records
StepFun product offering
Product offeringCore offering
StepFun develops and commercializes a series of multimodal foundation models (Step 1, Step 2, Step 3.5 Flash, Step 3.7 Flash, Step TTS, StepAudio 2.5 Realtime) that process and generate text, images, video, code, and speech. These models are delivered to enterprise developers through the Step Open Platform (开放平台) API and to end users through the Step AI (阶跃AI) consumer assistant available across desktop, mobile, and web.
Product overview
StepFun is a Shanghai-based AI company operating as a platform-plus-modules architecture centered on its Step series foundation models. The core products include Step 1 (efficient long-text language model), Step 2 (trillion-parameter deep reasoning model), Step 3.5 Flash (fast agent language model), and Step 3.7 Flash (flagship 198B MoE vision-language model for coding agents and search), complemented by Step TTS and StepAudio 2.5 Realtime for speech capabilities. These models are exposed through the 开放平台 (Open Platform) API for developers, the Studio platform for application building, and the 阶跃AI consumer chat product available across desktop and mobile. The company also offers Step Plan as a subscription tier for accessing the full Agent development workflow.
Differentiator
Problem solved
Functional benefit
Products and services
- Step 3.7 Flash A 198B-parameter sparse Mixture-of-Experts vision-language model with 11B active parameters per token, 256k context window, and native multimodal support through a 1.8B ViT encoder. Designed for real Agent workflows including coding agents and search workflows, with stable tool calling compatible with mainstream Agent frameworks. Targets enterprise developers and AI application builders.
- Step 3.5 Flash A fast language model optimized for Agent applications, emphasizing speed and efficiency in executing agentic tasks. Designed for AI developers building agent-based systems.
- Step 2 A trillion-parameter self-developed base model with deep reasoning capabilities aligned with frontier performance and precise multi-layer instruction following. Targeted at enterprise developers requiring high-end reasoning.
- Step 1 A language model optimized for low-bit single-GPU long-text inference, verified across tens of millions of users with efficient multi-turn instruction following capability. Suited for high-volume and cost-sensitive deployments.
- Step TTS A speech synthesis model for generating audio output from text inputs. Targeted at developers building voice-enabled AI applications.
- StepAudio 2.5 Realtime An end-to-end real-time speech large language model with customizable persona capabilities supporting Chinese and English via WebSocket API, featuring million-scale persona data augmentation, roleplay-specific RLHF alignment, and unified speech understanding and generation. Aimed at developers building real-time conversational voice AI applications.
- Step AI (阶跃AI) An AI chat assistant available across desktop (Mac, Windows), mobile (iOS, Android), and web platforms as a versatile work companion. Distributed as a freemium consumer product under the company.
- Step Open Platform (开放平台) API platform providing access to Step series models including language, speech, and multimodal models, with emphasis on stability, high performance, and easy integration for building AI applications. Offers subscription (Step Plan) and usage-based pricing to enterprise developers.
- Studio A platform for building AI applications by connecting models, tools, and workflows, aimed at developers exploring next-generation AI application development.
Quantifiable outcome
- SWE-Bench Pro score of 56.26% (up from 51.3% prior version)
- +3 more outcomes
Companies that use StepFun
Customer profileNamed customers5 records
Segments3 records
Ideal customer profiles3 records
StepFun technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
AI capability9 records
Feature5 records
StepFun partnerships and signals
Strategic signalPartnerships
Seven partnerships are on record, tiered minor, major and core.
- CAICT (China Academy of Information and Communications Technology)minorStepFun participated in CAICT's 'Token-based Cloud Services Quality Enhancement and Empowerment Assessment Program' alongside Alibaba Cloud, Huawei Cloud, JD.com Cloud, Lenovo, and others to establish industry standards for token-based cloud services.
- Huawei CloudmajorStepFun partnered with over 20 model providers in Huawei Cloud's AI Model Partner Program at INSPIRE 2026 conference. Models integrated into Huawei Cloud's agentic AI ecosystem supporting enterprise productivity use cases.
- DexmalmajorDexmal secured strategic financing from StepFun along with Zhipu AI, SenseTime, and Alibaba to accelerate scaling of embodied large models using Atomix's logistics data spanning over 20 countries. Combined entity plans to launch three-tier sorting system in June and debut DM0.5 model in July.
- Geely Auto GroupcoreGeely partnered with StepFun and Ecarx under its 'ecosystem co-creation' strategy to advance full-domain AI and build talent density in automotive. Geely VP Li Chuanhai outlined plans for AI integration from future cabins to autonomous driving.
- NVIDIAmajorNVIDIA provided enterprise-ready infrastructure for Step 3.7 Flash deployment with optimized frameworks including SGLang, TensorRT-LLM, and vLLM. NVIDIA NIM provides containerized inference microservices for production deployment with Day 0 fine-tuning via NVIDIA NeMo framework.
- Alipay (Ant Group)majorStepfun adopted Alipay's Token Pay platform for AI payment flows including token top-ups, subscription billing, and usage-based payments. Services embedded across smart glasses, in-car systems, AI assistants, and retail apps. Part of Alipay's agentic commerce ecosystem.
- Yin Qi (Megvii Co-Founder)coreYin Qi, co-founder of Megvii, appointed as chairman of StepFun to oversee the company's strategic direction and technical development. Represents integration of Megvii's computer vision expertise with StepFun's foundation models.
Scale indicators6 records
Recent moves9 records
Expansion highlights8 records
StepFun competitors and assessment
Company assessmentDirect peers
- DeepSeek: Hangzhou-based foundation model lab developing open-weight LLMs (DeepSeek-V3, R1) at highly competitive cost-efficiency benchmarks. Most direct competitor to StepFun in the Chinese open-weight foundation model category, sharing similar architecture innovations (MoE) and pricing pressure.
- Moonshot AI: Beijing-based foundation model developer behind Kimi, one of the most prominent members of China's 'Six Little Tigers' alongside StepFun. Competes head-to-head in long-context LLMs, consumer AI assistants, and enterprise API offerings with comparable funding scale.
- Zhipu AI (Z.ai): Beijing-headquartered AI startup developing the GLM series of foundation models, also among China's 'Six Little Tigers.' Competes with StepFun across LLM APIs, enterprise partnerships (including Dexmal co-investment), and government-backed commercialization pathways.
- MiniMax: Shanghai-based foundation model developer focused on large language and multimodal models, co-listed alongside StepFun in major Chinese AI funding rounds. Closely comparable in scale, geography, and product mix including API platforms and consumer applications.
- Baichuan Intelligence: Beijing-based foundation model startup developing open-source LLMs (Baichuan series). Competes directly in the Chinese open-weight model category, targeting similar enterprise and developer customer segments.
- 01.AI (Yi): Beijing-based AI startup led by Kai-Fu Lee developing the Yi series of open-source LLMs. Competes with StepFun in the Chinese open-weight foundation model market, targeting similar developer and enterprise customers with comparable multimodal capabilities.
Broad incumbents
- Alibaba (Qwen): Cloud and e-commerce giant whose Qwen foundation model family competes broadly with StepFun across LLM APIs, multimodal, and enterprise deployment via Alibaba Cloud. Represents the well-funded incumbent peer with integrated cloud distribution advantage.
- ByteDance (Doubao): Operates the Doubao foundation model series integrated into TikTok and broader consumer apps. Competes with StepFun in multimodal AI, consumer assistants, and enterprise APIs, with massive proprietary data advantages from ByteDance's social platforms.
- SenseTime: Hong Kong-listed AI company with deep computer vision and foundation model capabilities, co-invested alongside StepFun in Dexmal. Competes in multimodal vision-language models and enterprise AI deployment, with broader computer vision product portfolio.
Others
Market position
Strengths5 records
Weaknesses5 records
Competitive moat6 records
Key risks6 records
Key highlights7 records
Customer concentration
StepFun social profiles
Digital presenceStepFun financial estimates
Financial estimateRevenue estimate
Valuation estimate
StepFun leadership team
Management profileNumber of profiles
Profiles2 records
StepFun funding detail
Funding detailFunding overview
Funding rounds4 records
Investors17 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
StepFun M&A and investment
M&A and investmentM&A
Investments2 records
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about StepFun
What does StepFun do?
StepFun develops and commercializes a series of multimodal foundation models (Step 1, Step 2, Step 3.5 Flash, Step 3.7 Flash, Step TTS, StepAudio 2.5 Realtime) that process and generate text, images, video, code, and speech. These models are delivered to enterprise developers through the Step Open Platform (开放平台) API and to end users through the Step AI (阶跃AI) consumer assistant available across desktop, mobile, and web.
Is StepFun a public or private company?
StepFun is a private company. It is classified as venture growth investor backed and is currently operating.
When was StepFun founded?
StepFun was founded in 2023. It employs 251 to 500 people.
Where is StepFun based?
StepFun is headquartered in Xuhui, China, in the Asia region.
How does StepFun make money?
Three revenue lines are on record. Step API Platform is the primary driver. The others are consumer AI Applications and enterprise Licensing & Integration.
Who are StepFun's main competitors?
Direct peers on record are DeepSeek, Moonshot AI, Zhipu AI (Z.ai), MiniMax, Baichuan Intelligence and 01.AI (Yi). Broad incumbents are Alibaba (Qwen), ByteDance (Doubao) and SenseTime. OpenAI is listed as an others.
Does StepFun have an API?
Yes. StepFun offers the Step API platform providing access to its language models (Step 3.5 Flash, Step 2, Step 1), speech models (Step TTS), and multimodal/GUI models (Step 3.7 Flash). The platform emphasizes stability, high performance, and easy integration. Developers can access models through the开放平台 (Open Platform) at platform.stepfun.com. Developer documentation is at platform.stepfun.com.
What industry is StepFun in?
StepFun's product category is Foundation Models / AI Infrastructure. Its primary akta.pro industry code is HDAAACAA, Foundation Model Developers (LLM/Multimodal Model Labs), with a secondary code of HDAAACAG, Multimodal Generation (Text-Image-Video-Audio) & Creative Toolchains. Its NAICS code is 54151 and its SIC code is 7372.