getnews.online

AI

New models, tools and AI features you can use.

AIOfficial

Microsoft makes Execution Containers generally available to fence in AI agents

MXC lets developers set what files and networks an agent may use. The rules sit outside the agent, so it cannot give itself more access.

Why it matters. that run code need clear limits. MXC gives builders one policy format across Windows, macOS and Linux.

In simple words. Microsoft released a tool that sets limits for AI helpers that run code on a computer. It decides which files and internet connections a helper can use, and the helper cannot change those rules. It is for people who build these helpers.

For most peopleMake it about you →

Skip: nothing changes for you.

Your call: hype or real?

AIConfirmed

You can now check if an image, video or audio clip has a SynthID watermark

Google opened its SynthID detector to everyone. It also works inside the Gemini app and Chrome, and Apple is said to be joining soon.

Why it matters. A simple way to check media from partner AI tools. It only finds SynthID watermarks, so it cannot judge everything.

In simple words. Google lets anyone check if a picture, video or sound carries a hidden mark called SynthID, which some AI tools add. It only finds that one kind of mark, so it cannot tell you everything. It is handy for media from partner tools.

For most peopleMake it about you →

Try it: upload a file and see if it carries a watermark.

Your call: hype or real?

AI assistants: top picks right now

Compare →
Best for writing
ChatGPT$20 a month

Free: Yes, with limits on uploads, images, voice and deep research · GPT-6

GoodPeople who want one assistant that writes, codes, makes pictures and talks out loud.

Watch outThe best models and the highest limits are on the paid plans.

Best value
Gemini$20 a month

Free: Yes, with 3.6 Flash and varying access to 3.1 Pro · 3.6 Flash and 3.1 Pro

GoodPeople who already use Google apps and want a free assistant that also makes images and video.

Watch outSome features, such as Gemini 3 Pro in search, are only in select countries.

AI🇨🇳 From ChinaOfficial

Developers can now use DeepSeek's open math code on Huawei Ascend chips

DeepSeek released DeepGEMM-Ascend on September 30, a free matrix-math library for Huawei Ascend chips. Reports say it is one of six modules in the release.

Why it matters. Open tools make it easier to run AI on more kinds of chips. It matters most to developers who work with Ascend hardware.

In simple words. DeepSeek shared free computer code that helps AI programs run on Huawei's AI chips. Anyone can read and use it. It is mostly useful for people who build AI software.

For most peopleMake it about you →

Watch: more open tools help AI run on more kinds of chips.

Your call: hype or real?

AIOfficial

Google opens Playground, a tool that builds games from text prompts

Playground lets US adults describe a game in words, then play and share it. No coding is needed, and access depends on a Google AI plan.

Why it matters. It shows how far prompt-based creation is moving, but it is an experiment with limited access for now.

In simple words. Google has a new tool where you describe a game in words and it builds the game for you. You do not need to know coding. For now it is only for adults in the US, and what you can do depends on your Google plan.

For most peopleMake it about you →

Watch: a fun preview of making games by talking.

Your call: hype or real?

AIReport

Meta's Muse assistant now has its own iPad app

Muse arrives on iPad a month after its phone launch, with new work connectors like Asana, Canva, Figma and Zoom.

Why it matters. Meta is pushing Muse from a phone helper toward a task tool for work and shopping.

In simple words. Meta's AI helper, Muse, now has an iPad app, a month after the phone version. It can link to work tools like Asana, Canva and Zoom. If you try it, connect only the accounts you need.

For most peopleMake it about you →

Watch: a look at where assistants are heading.

Hype4
Real3
As advertised

Your call: hype or real?

AIConfirmed

ChatGPT gets GPT-6 and Intelligent UI: buttons, forms and charts in answers

OpenAI is rolling out GPT-6 in ChatGPT, with answers that can include tappable buttons, forms, calculators and interactive charts. Free and Go tiers follow on Oct 8.

Why it matters. This is the first big change to how ChatGPT answers look in a while, and it reaches free users within a day.

In simple words. ChatGPT is getting a new model, GPT-6, and answers that can show buttons, forms, charts and calculators, not only text. Paid users got it on October 7. Free users get it on October 8, so you can see if it helps.

For most peopleMake it about you →

Worth a look: it makes ChatGPT feel more like an app. Turn the visuals down in settings if you prefer text.

Your call: hype or real?

AIOfficial

Anthropic launches Claude Haiku 5.5, about 75% cheaper to run than Haiku 4.5

Anthropic's smallest model now starts at $0.10 per million input tokens and adds adjustable effort settings for the first time in a Haiku.

Why it matters. Cheap, fast models decide what AI features cost to run. Haiku 5.5 targets high-volume jobs like support chats and subagents.

In simple words. Anthropic released Claude Haiku 5.5, its smallest and cheapest AI model. It costs about 75% less to run than the one before it. That helps companies add AI to things like customer chat without spending much.

For most peopleMake it about you →

Skip: it is a cheaper small model. You may see it inside apps without noticing.

Your call: hype or real?

AIOfficial

Mistral previews Large 4, a 1T-parameter model with weights due end of October

Mistral's new flagship is in public preview on its API at $1.36 per million input tokens. The weights are promised for the end of October.

Why it matters. A European lab says a near-frontier model will be downloadable. The weights are not out yet, and the model is far too big for home PCs.

In simple words. Mistral, a European AI company, showed a very large new AI model that people can use online now. The company says you will be able to download it at the end of October. It is too big for a home computer, though.

For most peopleMake it about you →

Skip: interesting for open AI, but nothing to use yet.

Your call: hype or real?

AIOfficial

Anthropic opens a 3-tier program with fewer cyber blocks for verified defenders

Security teams can apply for Claude access with fewer cyber-related refusals. Project Glasswing members move into the top tier.

Why it matters. Claude refuses some security work by default. Verified defenders can now apply to get fewer blocks, which shows how labs gate risky skills.

In simple words. Claude sometimes refuses computer security work, because it could be misused. Anthropic now lets verified security teams apply for fewer of those refusals. It only matters if you do security work.

For most peopleMake it about you →

Skip: it affects security professionals, not everyday chats.

Your call: hype or real?

AIOfficial

Google unveils Gemini 4 Argon, but only vetted cyber defenders can use it

Google's new flagship can write up to 1 million tokens per reply. It is limited to the Fairwind Program; paid API users and AI Ultra come later, with no date.

Why it matters. It is Google's first flagship in about a year, but ordinary users cannot touch it yet, and outside tests are missing.

In simple words. Google showed a new, powerful AI called Gemini 4 Argon. For now only trusted security teams can use it. Everyone else has to wait, and Google has not said when.

For most peopleMake it about you →

Skip: a big model you cannot try. Check back once Google sets a date.

AIConfirmed

OpenAI's GPT-6.1 Sol nearly matches GPT-6 Astra at one-fifth of the price

The new model costs $2 per million input tokens and $10 output. It is live in ChatGPT Work, Codex and the API for paid plans.

Why it matters. Near-top coding and computer-use skill at a fifth of the price makes tools cheaper to run and to buy.

In simple words. OpenAI released a new AI model that, the company says, does almost as well as its best one but costs about one-fifth as much. That could make AI helper tools cheaper. People on paid plans can use it now.

For most peopleMake it about you →

Skip: it is a cheaper back-end model, not a new app.

AIConfirmed

OpenAI launches Dots, always-on agents, and a $500-a-month Pro tier

Dots are agents with their own cloud computer that keep working after you log off. They come first to Pro and Business Premium; a new Pro 500 plan adds faster models.

Why it matters. Always-on that act in 4,000+ apps are a big step. So is the price ladder, and the $200 plan now gets less.

In simple words. OpenAI made helpers called Dots that keep working on their own computer even after you log off. They come first to paid plans, and a new plan costs $500 a month. Wait for early reviews before you let one use your accounts.

For most peopleMake it about you →

Skip: agents that act alone need trust. Wait for real-user reports.

AIOfficial

Claude Sonnet 5.5 is 30% faster, up to 30% cheaper, and now runs free Claude

Anthropic's new mid-size model keeps Sonnet 5's price at $2 per million input tokens while using fewer tokens per task. The free claude.ai tier gets it.

Why it matters. Free Claude users get a stronger model, and developers get a cheaper way to run coding and office .

In simple words. Anthropic released a new AI helper called Claude Sonnet 5.5. It is faster and costs less for companies to run. People who use the free version of Claude get it too, so you can try it at no cost.

For most peopleMake it about you →

Worth a look: open claude.ai and see whether the free tier suits you.