In May, GOSIM Paris 2026 took over STATION F for two days of talks and workshops built around the agentic AI convergence — protocols, tooling, and the open-source builders actually shipping against them. The hallway conversations were the good part. People arrived to hear about agent frameworks and left having argued about them with the maintainers.

On October 16–17, that continues in Shenzhen.

Why Shenzhen

Shenzhen is where China’s software and hardware sit within driving distance of each other. It has the world’s densest electronics supply chain, a young population, and a founder culture impatient enough to have earned its own phrase — Shenzhen speed. For a conference about open source AI, the location is the argument: this is a place where a model, a board, and a working prototype can be in the same room by the end of the week.

That matters more now than it would have two years ago. The center of gravity for open-weight model releases has shifted, and a great deal of the recent momentum has come out of Chinese labs. If you have been pulling those weights down, fine-tuning them, and building products on top of them, you have been working with the output of teams most people outside the region have never heard speak.

The people behind the open models

The Open Source Models track is the direct answer to that. It follows what is actually landing across language, vision, audio, and video — the breakthroughs, the deployments that survived contact with production, and the capabilities emerging from the open model ecosystem rather than behind an API.

The value of being in the room is the part that does not travel through a model card. Why a training decision was made. What got cut. Which benchmark the team themselves does not trust. That is the conversation Shenzhen is set up for, and it is much harder to have from eight time zones away.

That room is not hypothetical. Across the last three GOSIM conferences — Beijing 2024, Hangzhou 2025, and Paris 2026 — the programme has included Zhipu (now Z.ai) at every one, Moonshot AI (the team behind Kimi) at two, DeepSeek’s heads of data and AI governance, engineers from Alibaba’s Qwen team and Tongyi Lab, ModelBest of MiniCPM, and BAAI — its chairman and chief engineer among them — plus researchers from Shanghai AI Laboratory. With them were the people maintaining what actually runs these models: SGLang and vLLM on the server, and llama.cpp on your own machine — its core maintainer among several Hugging Face engineers on the programme. That is the other half of the conversation: not just how a model was trained, but what it takes to run it. Shenzhen’s speaker list is still being assembled, but that is the room GOSIM has been building.

Six tracks

The program runs across six tracks:

  • Agentic AI Summit — the strategy layer of the agentic wave: protocols, software patterns, tooling, and business models, including MCP, A2A, sovereign AI, and AI-native development.
  • Agentic AI on Edge — on-device inference acceleration, model compression, and real-time decision-making in resource-constrained environments.
  • Agentic OS & App — the operating system layer emerging to orchestrate agents: multi-agent coordination, runtime environments, intent-driven apps, the future of what user interfaces will become.
  • Open Source Models — the latest breakthroughs and practical deployments across the open model ecosystem.
  • Open Source Robotics — where agentic software meets physical systems: open robotics platforms, embodied AI stacks, autonomous machines that perceive, plan, and act.
  • AI Hardware — accelerators, chips, systems, and open architectures built to make training and inference faster and more accessible.

Alongside the talks, GOSIM Spotlight puts the most promising open-source AI projects on stage for live, hardcore demos, pitching directly to global investors and technical audiences — plus hackathons and a robotics challenge for people who would rather build than watch.

Getting there

The conference will be at the ADEN Hotel Shenzhen Nanshan in Shekou, about 700 metres from Sea World Metro Station on Lines 2, 8 and 12.

The call for proposals is open now. If you maintain an open-source AI project, have shipped something on top of open weights, or have a result worth arguing about, this is the moment to put it forward — speakers and the full schedule are still being announced.

Everything else lives at shenzhen2026.gosim.org. Please visit for details and feel free to contact us for any questions.