// this week in AI

This week in AI: GPT-6 free in ChatGPT, Gemini 4 Argon, and a Google agent with its own Gmail inbox

By Trinity · MariVisionOctober 16, 20267 min read
Ale from MariVision in a large dark newsroom, under a curved wall of holographic screens showing the number six, a tube of violet light, a glowing hummingbird, an envelope with a small robot, and a spiral
2026 · 10 · 16 — THIS WEEK IN AINew models for everyone, prices down, AI becomes a coworker

A week of new models, and not just for paying users. OpenAI brings GPT-6 into ChatGPT even on the free plan, Google answers with Gemini 4 Argon and unveils a work agent with its own email inbox, Anthropic slashes the price of its smallest model, and Meta releases the engine behind its Muse agent.

As every Friday, for each story you'll find what happened and why it matters if you work for yourself or run a small business, with sources at the end.

1 · GPT-6 lands in ChatGPT, even on the free plan

Large dark office with rows of empty desks and a holographic chat window floating near the ceiling, with a bar chart, a calculator keypad, sliders, and a big glowing number six

On October 7, OpenAI started rolling out GPT-6 in ChatGPT. If you're on a Plus, Pro, Business, or Enterprise plan, you'll find GPT-6 Sol as the new default model; since October 8, Free and Go users get GPT-6 Luna, the lighter version. The rollout is gradual, so you might see it a day after your coworker does.

The most visible change is Intelligent UI: when it helps, the answer isn't just text but includes elements you can use right there, like charts, small calculators, maps, or a form to split an expense. You can turn it off in the web version's settings. According to OpenAI, the fast version of GPT-6 answers web-search questions 44% faster than the previous one. The models behind Codex and the Work section don't change, and if you want, you can switch back to GPT-5.6 to compare.

Why it matters to you: if you use ChatGPT for quotes, margins, or quick math, you can now ask for a calculator with sliders instead of a wall of numbers, even without a subscription. Handy for thinking things through, but the numbers that end up in a client offer are still yours to double-check, same as before. If you want to figure out which tasks are worth handing to AI in your business, you'll find examples on the page about AI consulting for SMEs.

2 · Google launches Gemini 4 Argon, its new flagship model

Towering dark lab seen from afar, with a large holographic glass tube full of violet light floating near the ceiling, next to a price tag and a calendar page

On October 13, Google unveiled Gemini 4 Argon, its new frontier model. In the Gemini app and in AI Mode in Search, it's available to Google AI Pro and Ultra subscribers, not on the free plan. Developers can find it in AI Studio, businesses in Gemini for business.

According to Google, it ranks first on the Vals Index, a leaderboard that measures work in finance, coding, legal, and tax, and via API it can read up to a million tokens in one go. The launch API price is $2 per million input tokens and $10 per million output tokens, until November 30; starting December 1, it goes up to $4 and $20.

Why it matters to you: if you already have Google AI Pro, you'll find it in the app without lifting a finger. If you have automation that uses Gemini via API, now's the time to try it, but run the numbers on the full price: a workflow that pencils out at the launch price may not anymore in December, when it costs twice as much.

3 · Claude Haiku 5.5: the small model costs up to 90% less

Large dark mail sorting room with shelves full of envelopes and a glowing holographic hummingbird floating near the ceiling, sorting a stream of envelopes, next to a price tag with a downward arrow

On October 7, Anthropic launched Claude Haiku 5.5, its smallest and fastest model. It costs $0.10 per million input tokens and $0.50 per million output tokens for requests up to 100,000 tokens, versus $1 and $5 for Haiku 4.5: 90% less on short texts, 50% less on long ones, and, according to Anthropic, about 75% less on average. It's also the first Haiku where you can adjust how much it “thinks” before answering.

Anthropic positions it for repetitive, high-volume work: summaries, classification, querying a database, live customer support. It also says plainly that on complex programming tasks it still trails Sonnet 5.5 and Opus 5.5. In the same announcement, it cut the price of Sonnet 5.5 cache reads in half.

Why it matters to you: plenty of useful automation never got off the ground because the monthly bill didn't add up: reading every incoming email, sorting requests, summarizing client documents. At a tenth of the cost, jobs that didn't pay off yesterday pay off today. That's exactly the kind of workflow we build in our AI automation for businesses, starting from the tools you already use.

4 · Google unveils the Gemini agent for work, with its own Gmail inbox

Large dark office with empty desks and a holographic company ID badge with a small robot floating near the ceiling, surrounded by an envelope, a calendar page, and a gauge with a red limit mark

On October 8, at the Gemini at Work event, Google unveiled a single Gemini agent for work, built into Gemini Enterprise at no extra cost where the service is available. It works from the web, your phone, your computer, and even inside Microsoft 365 and Slack, and it can use both Google's models and Claude's.

The interesting part is how it's treated: each agent has its own identity, with limited access, and everything it does is logged under its name. A permanent agent, a kind of coworker, can have its own Gmail address, a calendar, and space for files. Admins set a spending cap per project: when the agent hits it, it stops until someone restarts it. For now, access is limited to a small number of companies.

Why it matters to you: you won't be using it tomorrow, but it's the model that will reach small businesses too: AI as a coworker with a name, an inbox, and a log of what it does. Two rules you can copy right away, with any tool: give the agent only the access it needs and set a spending cap. We cover this on the page about AI agents for businesses.

5 · Meta releases Muse Wonder, the engine behind its agent

Towering dark library seen from afar with endless shelves and a large holographic spiral of cards and photos floating near the ceiling around a glowing infinity symbol, with small social media icons around it

On October 12, Meta's superintelligence lab released Muse Wonder, the company's first frontier model, along with a larger version, Muse Wonder Full. According to Meta's tests, its scores are comparable to OpenAI's and Anthropic's flagship models, and the point Meta keeps stressing is “active memory”: the ability to remember and connect users' information.

Via API it costs $4 per million input tokens and $20 per million output tokens; the Full version is reserved for Muse Code subscribers. Both models will power the Muse agent, the same one Meta started bringing to small businesses at the end of September, connected to Instagram and Facebook pages and to tools like Shopify and QuickBooks.

Why it matters to you: if you've tried Muse for your business or are thinking about it, the engine is getting much more powerful and, above all, remembers more. Before connecting it to your accounts, take your time deciding what it should know and what it shouldn't: customer data, confidential pricing, and private conversations shouldn't be fed to an assistant just because it's convenient.

In short

GPT-6 comes to ChatGPT even on the free plan, with answers that include calculators and charts: useful, but numbers for clients still get double-checked. Gemini 4 Argon is for Google AI Pro or Ultra subscribers, and via API it costs half only until November 30. Claude Haiku 5.5 makes repetitive automation that used to cost too much worth it. The Gemini agent for work shows what AI at the office will look like, with its own identity and a spending cap. And Meta gives its Muse agent an engine with far more memory, so it pays to decide up front what you let it remember.

Sources: Android Headlines (GPT-6 in ChatGPT) · OpenAI Help Center (GPT-6 and other models in ChatGPT) · Google (Gemini 4 Argon) · Anthropic (Claude Haiku 5.5) · Reworked (Gemini agent for work) · TechCrunch (Meta Muse Wonder) · TechCrunch (Muse for small businesses)

Frequently asked questions

Is GPT-6 available on the free version of ChatGPT?

Yes. Since October 8, Free and Go users get GPT-6 Luna, the lighter version, while Plus, Pro, Business, and Enterprise plans use GPT-6 Sol. The rollout is gradual and can vary from one account to another. Interactive answers can be turned off in the web settings.

How much does it cost to use Gemini 4 Argon in automation?

Via API, Google offers it at $2 per million input tokens and $10 per million output tokens until November 30, 2026. After that, the price doubles to $4 and $20. In the Gemini app, it's included only with Google AI Pro and Ultra subscriptions, not on the free plan.

Should you use Claude Haiku 5.5 instead of a bigger model?

For repetitive work, yes: sorting email, summarizing documents, classifying requests, answering frequently asked questions. On short texts it costs 90% less than Haiku 4.5. For complex reasoning and demanding programming, Anthropic itself points to Sonnet 5.5 and Opus 5.5 as the better fit.

Want to know which of these updates actually matter for your business, and which you can ignore?
A half-hour chat, no commitment, plain talk, no jargon
Let's talk →