Hong Kong AI Blog
What’s actually happening with AI in Hong Kong — the tools, the models, the people, and the work.
The Hong Kong AI Wall: Which US AI Tools Block HK and Why
OpenAI, Anthropic, and Google all block Hong Kong from their AI services. Here's exactly what's blocked, what still works, and what HK developers are doing about it.
The HK Developer's AI Toolkit: What Actually Works Without a VPN
A practical guide to AI tools that work in Hong Kong without a VPN — from coding assistants to LLM APIs to self-hosted models.
Cursor in Hong Kong: Why It's the Go-To AI Coding IDE
Cursor works in Hong Kong without a VPN. Here's how Auto mode, model selection, and the Pro plan work — and what to watch out for.
OpenCode: The Open-Source Coding CLI That Works Everywhere
OpenCode is a free, open-source terminal coding assistant with MiniMax M2.5 included free. Works with any OpenAI-compatible API. No VPN required.
The Chinese AI Model Guide: DeepSeek, Qwen, Ernie, and Beyond
A comprehensive English-language guide to Chinese AI models — who makes them, what they can do, whether they're open source, and how to use them from Hong Kong.
StepFun 3.5 Flash: 11B Active Parameters That Punch Above Their Weight
StepFun's 3.5 Flash is a 196B MoE model that runs with only 11B active parameters — scoring 97.3% on AIME while being fast and cheap. Here's why HK developers should pay attention.
Qwen 3.5: Alibaba's Answer to GPT — Open Source in 201 Languages
Qwen 3.5 supports 201 languages, offers models from phone-sized to 397B MoE, and it's all Apache 2.0. For bilingual HK developers, this might be the best model available.
MiniMax M2.5: Claude Opus Performance at 1/20th the Cost
MiniMax M2.5 matches Claude Opus on benchmarks at a fraction of the price, and it's open source. Plus, the company just IPO'd on HKEX.
From Kimi to GLM-5: The Chinese AI Companies That IPO'd in Hong Kong
Zhipu, MiniMax, and StepFun are all listing on HKEX. Hong Kong is becoming the financial capital of Chinese AI.
Hong Kong's AI Ecosystem: 300 Companies You Didn't Know Existed
Hong Kong has over 300 AI companies, world-class research labs, billions in government funding, and a 3,000 petaFLOPS supercomputing centre. Here's the landscape.
HKGAI and HKChat: Hong Kong's Homegrown LLM
Hong Kong built its own LLM. HKGAI fine-tuned DeepSeek for local needs and launched HKChat — a public AI chatbot for HK citizens. Here's how it happened.
HK$1 Billion for AI: What AIRDI Means for Hong Kong's Tech Future
Hong Kong approved HK$1 billion for a dedicated AI Research & Development Institute, plus billions more for frontier tech. Here's what's planned and why it matters.
The Research You're Missing: AI Papers from HKUST, HKU, and CUHK
Hong Kong's universities are publishing world-class AI research — from 100K-agent simulations to autonomous AI scientists. Here's what's coming out of HK's labs.
OpenClaw Meets Chinese AI: Building Agents in Hong Kong
OpenClaw is an open-source AI agent framework that works with DeepSeek, GLM-5, and MiniMax. Here's how it can be used to build autonomous agents in Hong Kong.
Microsoft Copilot: The Compliant Loophole for HK Companies
Microsoft Copilot uses OpenAI's models but is officially available in Hong Kong. For regulated industries like finance and legal, it's the most compliant path to GPT-4. Here's what you get, what you give up, and whether the models lag behind.
Self-Hosting DeepSeek in Hong Kong: A Practical Guide
How to run DeepSeek models on your own hardware in Hong Kong — from Ollama on a MacBook to vLLM on cloud GPUs. A step-by-step guide for different budgets.
Switching from ChatGPT to DeepSeek: What Changes, What Doesn't
A practical migration guide for Hong Kong developers moving from ChatGPT to DeepSeek — API compatibility, prompt differences, and what to expect.
vLLM vs LM Studio vs llama.cpp: How to Self-Host AI Models in Hong Kong
Three ways to run AI models on your own hardware. Here's when to use vLLM (production), LM Studio (desktop), or llama.cpp (everywhere else).
Why Qwen3-235B-A22B Is So Good: A Technical Deep Dive
Qwen3-235B-A22B is Alibaba's flagship MoE model — 235B total params, 22B active, competitive with GPT-4o and Claude. Here's the architecture and why it matters for Hong Kong developers.
How to Get Gemini in Hong Kong: Vertex AI, Workspace, and the Gotchas
Google's Gemini isn't directly available in Hong Kong — but there are official paths. Here's how to access it through Vertex AI and Google Workspace, what's actually blocked, and the hidden landmines you'll hit.
Poe in Hong Kong: Access Claude, GPT, and Gemini Without a VPN
Poe by Quora is an AI aggregator that gives Hong Kong users access to Claude, GPT-4, Gemini, and 100+ other models through a single app — no VPN required. Here's what it actually is and when to use it.