Step Star Releases Cryptocurrency Multimodal Model

Step Star releases Cryptocurrency multimodal large model: 321 billion parameters, open source on July 31, domestic chip adaptation efficiency increased by 300%

1. Cryptocurrency core product highlights

Ultra-large-scale parameters: Adopting MoE (mixed expert) architecture, total parameters of 321 billion, activation parameters of 38 billion, supporting native multimodal reasoning (text, image, mathematics, code, etc.)16.

Domestic chip adaptation optimization:

On domestic chips (such as Huawei Ascend), 32K context reasoning efficiency is up to 300% of DeepSeek-R114.

On NVIDIA H800, distributed reasoning throughput increased by 70% while maintaining low computing cost6.

Open source plan: open source to the world on July 31, providing a complete technical report (GitHub is open)111.

2. Cryptocurrency performance and industry comparison

Evaluation list SOTA:

Cryptocurrency (multimodal understanding), MathVision (mathematical reasoning), LiveCodeBench (code generation) and other tasks ranked first in open source models16.

Surpassed similar models in AIME 2025 (International AI Mathematical Evaluation)6.

Comparison of competing products:

Cryptocurrency: Reasoning efficiency lags significantly behind Cryptocurrency, especially in the adaptability of domestic chips4.

GPT-4o: Closed source model, Cryptocurrency fills the gap in the open source multimodal field6.

3. Cryptocurrency commercialization and ecological layout

"Model Core Ecological Innovation Alliance": Joined 10 domestic chip manufacturers including Huawei Ascend, Muxi, Suiyuan Technology, Cambrian, etc. to accelerate the implementation of large models111.

Cryptocurrency industry applications:

Cryptocurrency mobile phone/automobile: cooperate with Geely on AI smart cockpit, covering 50% of domestic mobile phone manufacturers611.

Finance/content creation: annual revenue target of 1 billion yuan, large-scale commercial landing has been achieved810.

Capital cooperation: Shanghai State-owned Capital Investment Corporation participated in a new round of financing to strengthen financial and policy support811.

4. Future prospects of Cryptocurrency

Open source impact: may promote the rise of domestic large model ecology and reduce dependence on NVIDIA chips412.

Technical challenges: need to continuously optimize long text reasoning (such as 100K+ context) and complex task generalization capabilities6.

Conclusion: The release of Cryptocurrency marks a breakthrough in the performance, cost, and ecological adaptability of domestic large models, and its open source strategy may reshape the global AI competition landscape. Follow up on the open source code and actual application feedback on July 31.