Stepfun Open-Source Small Model Ecosystem & IPO Capital Operations

📌 Background

In 2026, the divergence among China's 'Six Little Tigers' of large language models accelerated. Following the Hong Kong Stock Exchange listings of Zhipu and MiniMax, Stepfun completed a Series B+ financing round exceeding 5 billion RMB and initiated its IPO in Hong Kong, targeting a valuation of over 10 billion USD. Driven by the open-source impact of DeepSeek, Stepfun shifted from a closed-source to an open-source strategy, releasing Step 3.5/3.7 Flash focusing on inference efficiency and low cost.

👤 Target Customers

Paying customers include enterprises and developers using the open platform API, government and enterprise clients purchasing industry solutions, and AI application teams requiring lightweight models for secondary development.

💰 Revenue Streams

1) Open platform API billing based on token usage; 2) Open-source Flash series driving inbound traffic to monetize flagship closed-source models and cloud services; 3) Fees from industry solutions and customized deployment projects.

🧮 Cost Structure

Computing power investments for trillion-parameter MoE training represent the largest expense, coupled with inference cluster operations, R&D personnel salaries, and IPO compliance costs.

🛡️ Moat

Inference efficiency and cost advantages under the trillion-parameter MoE architecture form a technical barrier; the open-source strategy fosters a thriving developer community; Shanghai state-owned asset backing and cash reserves exceeding 5 billion RMB support long-term operations.

🔑 Keys to Success

• Continuous iteration of open-source small models to maintain developer stickiness
• Leveraging inference cost advantages to secure benchmark enterprise-level large clients
• Validating the commercial revenue structure prior to the IPO

⚠️ Risks

• Expanding losses putting pressure on IPO valuation
• Flagship models facing a dual squeeze from closed-source competitors and free open-source models

🏢 Cases

• Completed a Series B+ financing round exceeding 5 billion RMB in January 2026, setting a record for a single financing round in China's LLM sector
• Step 3.5-Flash ranked at the top of multiple benchmarks for inference efficiency, garnering attention from overseas developer communities

📊 SWOT Analysis