A whole country pays to make AI practical

Wednesday, September 16, 2026

The most interesting thing in AI today is not a new model, it is who gets to learn and build with one. Singapore rolled out more than 200 AI training courses and, for eligible citizens who enroll and attend, six months of free access to premium tools such as ChatGPT Plus and Google AI Pro. Google shipped Gemini 3.8 Live, a voice model that can reason and call tools while it listens, priced well under the main rival. And OpenAI started testing Sponsored Agents in ChatGPT, where clicking an ad can open a conversation with a business before you decide. The theme is access. Cheaper models, free training, and a race to make AI useful to regular people, not just the biggest labs.

Today's top stories

Channel NewsAsia

A whole country funds AI skills and hands out free premium tools

Channel NewsAsia reports that SkillsFuture, Singapore's workforce agency, rolled out more than 200 AI training courses from about 30 providers. Eligible citizens aged 18 and up who enroll and attend get six months of free access to premium AI tools, with choices that include ChatGPT Plus, Google AI Pro, Manus AI, and Microsoft 365 Personal.

Why it matters The biggest barrier most people hit is not finding a course, it is paying for the tools to practice with after class. Singapore removes that barrier and pairs it with a careers center and an AI-readiness quiz. Anywhere you are, look for a government, employer, or library program that already covers tool access, and budget the tool alongside the course.

OpenAI

ChatGPT is testing ads you can talk to

OpenAI is testing Sponsored Agents, which let you click an ad in ChatGPT and start a conversation with an AI representing the business. You can ask follow-up questions and follow a link to the business's website. The test covers select advertisers in the United States, and OpenAI says these conversations are clearly labeled and separate from your original chat.

Why it matters You can ask about dimensions, delivery, or which version fits your needs without hunting across product pages. But the conversation is with a seller's representative, not an independent expert. Ask questions, then verify the important details on the business's website and compare alternatives separately before you decide.

The Decoder

Google ships a cheaper voice model that reasons while it listens

The Decoder reports that Google DeepMind released Gemini 3.8 Live and 3.8 Live Extended Thinking, two audio models for the Gemini API and AI Studio. The Live version powers voice agents that can make API calls in the background, process what they see, and keep talking, in more than 97 languages. Google charges about $0.005 per minute for audio input and $0.018 for output, roughly $1.38 for an hour of voice conversation.

Why it matters If you have wanted to build a voice assistant or add talking to a small app, the cost has stopped being the blocker. Try the sample apps on GitHub or test it in Google AI Studio with one real task. Pricing is early, so recheck before you scale anything.

More AI Signals

OpenAI Help
ChatGPT tasks can now start automatically from Gmail, Slack, or GitHub
Instead of waiting for a set time, a ChatGPT task can run the moment a new email arrives, a Slack channel gets a message, or a pull request changes. It is available on Plus and higher plans, and it turns plain-English triggers into background automation anyone can set up.

LLM Releases and New Features

Google DeepMind
Gemini 3.8 Live and 3.8 Live Extended Thinking hit the API
These are audio models that think and act while you talk, with a price undercut that is the real story. Builders testing voice should try them this week.
NVIDIA
NVIDIA open-sources SoL-Pi, a tool that cuts coding-agent token use
SoL-Pi is a free, open-source extension for the Pi coding agent. It cuts token use by roughly 45 to 49 percent and about one-third of the cost on EdgeBench while keeping most of the task score, using methods that stop the agent from repeating old context and reading giant logs.

Tools and Workflows

NVIDIA
NVIDIA's SoL-Pi makes a coding agent spend less
SoL-Pi is opt-in and free, so install it and A/B it on one longer task. It suits builders who keep hitting token limits, and it needs the Pi coding agent to run.

New in Robotics

Xiaomi Robotics
Xiaomi open-sources a robot brain that can picture its world
Xiaomi released Xiaomi-Robotics-U0, an open-source 38 billion parameter model that ties together image generation, scene editing, and embodied video for robots. It is designed to let a robot imagine and plan around its environment before it moves, with open code and checkpoints available to test.

Courses and Learning

University of Hawai'i News
AI for Hawai'i is a free, open course for everyone
The University of Hawai'i's AI for Hawai'i is a free, self-paced course open to anyone, with no university login required. Its 12 chapters run from how AI works through responsible use, strategy, and practical workplace skills. Start with the first chapter and do one a week.

AI Jobs

Anthropic
Anthropic is hiring a senior AI engineer to build its own agents
Anthropic lists an AI Engineer for GTM Claudification, a senior role building the autonomous agents that run its go-to-market work, from MCP servers to evaluation frameworks. It is remote-friendly but travel-required and based out of San Francisco or Seattle, so verify the location and logistics in the posting before applying.

Events

AI Insights Summit
AI Insights Summit 2026 starts tomorrow
AI Insights Summit 2026 runs September 17 through 19, free and fully virtual, with more than 40 experts across business, health care, education, and technology. Registration is still open, and no coding background is needed.

Back to daily archive Previous issue 2026-09-15