Your Next AI Project Inspiration Lies in the Seven Forums of the First MoDa Developer Conference}

What kind of era are we living in? It’s the "Second Half of AI" envisioned by Yao Shunyu, the "Software 3.0 Era" defined by Andrej Karpathy, a time of creator-AI co-evolution, and a redefinition of "possibility" itself.

Innovative models are emerging at an unprecedented pace, profoundly transforming every corner of the world. For developers, this is a golden age of opportunity but also a challenge: how to stay at the forefront, efficiently access, utilize, and create AI models?

The community ecosystem of openness, collaboration, and sharing has become the core driver of the AI wave. In this context, a platform that gathers top wisdom, offers comprehensive support, and connects creators with users is crucial. ModelScope is precisely such a platform.

On June 30, the first MoDa Developer Conference was held in Beijing. Since its founding in November 2022, the community has rapidly grown, now hosting over 500 contributing organizations, with more than 70,000 open-source models—an increase of over 200 times. User numbers have expanded from 1 million in April 2023 to 16 million now, a 16-fold growth.

ModelScope now offers full-chain services supporting developers in experiencing, downloading, tuning, training, inference, and deploying models across fields like LLMs, dialogue, speech, image generation, video synthesis, and AI composition, with over 4000 MCP services and debugging tools. It has become China’s largest AI open-source community, with leading models first open-sourced here.

By aggregating cutting-edge open-source models, ModelScope enables developers to quickly access the latest and best models, while also providing a bridge to potential users and downstream ecosystems. The mutual engagement of model contributors and users sparks endless application possibilities.

Seven Forums Unlock the Latest AI Trends

ModelScope is an open, neutral, non-profit organization. The conference, guided by the National Information Center, features a main forum plus six thematic forums covering 65 topics, with renowned AI open-source teams sharing insights on cutting-edge models and tools.

From these forums, we observe new trends in AI development.

Open Source

In 2025, the global AI open-source wave surges, with China becoming a key and unique driver. Companies like Alibaba (Tongyi Qianwen) and DeepSeek continue releasing top-tier open-source models. The Qwen series is now a preferred tool worldwide, fostering innovation from academia to industry.

These models are pathways to technological independence and self-reliance, forming a robust AI ecosystem. This includes vibrant developer communities like ModelScope and closer integration with national infrastructure, driving AI applications in public services and manufacturing.

Multimodal and World Models

AI no longer just processes text or images. The development of multimodal AI now enables understanding and generating text, images, audio, video, and even 3D signals, allowing more natural and comprehensive interactions. Examples include GPT-4o, which produces hyper-realistic images, and Veo 3, capable of stunning video synthesis, making the virtual seem indistinguishable from reality.

Closely related is the rise of world models, where AI begins to internalize the physical laws of the world, enabling reasoning, prediction of physical interactions, and better understanding of human intentions. Video generation models now do more than create visuals—they understand causal relationships, laying the foundation for robotics, autonomous driving, and advanced virtual assistants.

Small Models and Edge Applications

While pursuing larger, more powerful models, the industry increasingly focuses on efficiency and cost. Large models are expensive to run in the cloud, prompting a shift toward model compression, quantization, and distillation. This leads to powerful yet compact "edge AI" models that run directly on personal computers, smartphones, or IoT devices, reducing latency, dependency on cloud services, and better protecting user privacy.

At the conference, Tsinghua University’s Dean of Electronic Engineering, Wang Yu, shared breakthroughs and challenges in hardware innovation and edge AI, introducing the open-source multimodal small model Megrez-3B.

Embodied Intelligence

If multimodal AI gives AI "five senses," embodied intelligence gives AI a "body." In 2025, breakthroughs in combining AI with robotics are emerging. Humanoid robots equipped with advanced visual-language models are stepping out of labs to perform complex tasks like warehouse sorting and home assistance in unstructured environments.

The key is not just hardware progress but the synergy between "brain" and "body." AI models need to convert multimodal perception into physical actions in real-time, learning and adapting through interaction.

Tsinghua’s researcher Zhao Xing, co-founder of Xinghai Tu, discussed building an ecosystem for embodied AI developers, focusing on key elements like ontology, data, models, and applications, sharing insights on fostering this ecosystem.

Agent and MCP

In 2025, autonomous AI agents capable of understanding, planning, and executing complex tasks are a hot topic. Ensuring these agents are controllable and reliable is crucial for large-scale deployment.

New interaction paradigms are being explored to transform agents from unpredictable "black boxes" into transparent, controllable partners. The Model Context Protocol (MCP) framework standardizes communication, allowing agents to clarify goals, demonstrate plans, and request authorization before acting—ensuring human control and a leap from "usable" to "trustworthy" AI.

Deep Applications of Generative AI

Beyond text and image generation, 2025 sees generative AI making significant impacts in:

• Scientific Discovery: Accelerating new material discovery, drug design, and complex simulations.
• Engineering & Design: Generating and optimizing complex 3D models, circuits, and industrial processes.
• Software Development: Not just coding snippets but understanding entire codebases for refactoring, debugging, and documentation, becoming developers’ "pair programmers."
• Personalized Content & Entertainment: Real-time interactive virtual worlds and games, with AI deeply integrated into film and TV production.

Developer Incentive Program

The conference announced a developer badge incentive program, rewarding contributors with honors, free GPU compute, advanced training vouchers, and image generation credits. The program aims to foster community growth, encouraging developers to share models, collaborate, and innovate, driving the next wave of AI technology.