AI tools & everyday work

AI Updates: What to Use for What in October 2026

October 28, 20268 min readAlice Boone

You sit down to write a product description. Twenty minutes later, you're comparing model names and still haven't written a sentence. The useful question is smaller: what are you trying to finish?

Here is how I'd match the current tools to everyday jobs, followed by the recent updates worth knowing about. This is the October 28, 2026 snapshot, with links to the original announcements and documentation.

What would you like to get done?

You do not need a favorite model for everything. You need a useful starting point for the work in front of you.

On a smaller screen, scroll sideways to read all three columns.

A starting point by task
The job	Where I'd start	Give it something specific
Writing or editing	Claude Sonnet 5.5; GPT-6.1 Terra in ChatGPT Work or Codex	The real notes, audience, length and a sample of the voice you want
A complicated brief or document	GPT-6.1 Sol in Work/Codex, Claude Opus 5.5 or Gemini 4 Argon	The files, the question to answer and what the finished result should contain
Research across sources	Gemini Deep Research	A narrow question, a date range and a request for links supporting the important claims
Building or fixing a website	Codex or Claude Code	The actual project, the problem and a clear way to check the result
Making or changing an image	ChatGPT image generation or Gemini's Nano Banana	A reference image and a description of what should change and stay the same
Creating a short video	Gemini Omni	A short scene, the movement you want and reference media when useful
Lots of small, repeatable jobs	Claude Haiku 5.5 or a Luna model in Work/Codex	A few examples, a consistent output format and a way to spot mistakes

These are practical starting points based on the capabilities described in the sources below. For your own work, compare the result you can actually use: the edited paragraph, working page or finished file.

The model is the engine. The app supplies the tools around it. For editing files or checking a website, those tools matter as much as the model name.

What changed recently?

October 28: Google announced Gemini 4 Carbon

Carbon's announcement focuses on difficult engineering work and cybersecurity. Its initial rollout is to selected cyber defenders and trusted partners. A general consumer release is still ahead.

For someone choosing a model in the Gemini app today, Gemini 4 Argon is the more relevant recent release. Google began rolling it out to AI Pro and Ultra subscribers on October 13 for complex tasks such as analysis, presentations and interactive projects.

This week: GPT-6.1 Terra joined Work and Codex

OpenAI's October 26–30 update introduces Terra in ChatGPT Work and Codex. Its focus is understanding images and screens, using computer tools and writing clearly. That makes it interesting for jobs such as reviewing design assets or working through a defined visual task. This availability applies to Work and Codex, rather than the ordinary Chat model picker.

Ongoing work is becoming part of the assistant

On October 21, Anthropic opened a waitlist for Claude Duet, starting with US Pro and Max subscribers. The research preview is built around an assistant that can keep working between conversations and bring things back to your attention.

OpenAI's update this week also includes texting a dot through iMessage or Android RCS, in beta for Pro users in the US. These developments are about how an assistant fits into your day, as well as the answer it gives in one conversation.

Smaller models are getting useful work to do

Anthropic launched Claude Haiku 5.5 on October 7, with an emphasis on fast, high-volume tasks such as summaries, sorting and routing. Think grouping a batch of support messages or putting short descriptions into a consistent format.

That is useful progress. A repetitive task can benefit more from a clear process and a fast model than from sending everything to the largest one.

If you're writing, bring the real material

For a product page, newsletter or first draft, I'd start with a tool that makes revision easy. Ask it to work from your actual details, then tell it what feels wrong: too formal, too vague, too long, or a claim you would never make.

Anthropic positions Sonnet 5.5 around well-defined everyday work, including documents and code. Its release notes also describe clearer writing. OpenAI lists Terra for writing and visual work, and Sol for more involved projects. Those are sensible places to start, with the model chosen inside the product you're using.

A useful brief might say: “Write for someone who likes the product but needs to know the size, file format and what happens after checkout.” That gives the draft a job. “Make this amazing” mostly gives it permission to add adjectives.

If you need reliable answers, ask for the evidence

A research task starts with a question small enough to answer. “Tell me about selling printables” is enormous. “Compare the export options described in these two help pages” gives you something you can check.

Gemini Deep Research can work through web sources and, when you choose them, your files and Workspace content. I'd use it for collecting and comparing information, then open the sources behind any claim that affects the work.

Ask for a separation between what a source says and what the assistant is inferring. A link at the end of a paragraph is useful only if the linked page supports the paragraph.

If you're building something, work in the project

For a website change, a coding assistant that can read the files and check the rendered page is useful. Give it the existing project and a bounded request: which page, which behavior, what needs to stay, and how you will know it worked.

For example: “Make the mobile menu usable with a keyboard. Keep the current layout and colors. Check opening, closing and focus.”

Codex and Claude Code are starting points for that kind of work. Within those tools, OpenAI's model guide places Sol around complex work and Luna around focused, repeatable jobs. Anthropic describes Sonnet 5.5 as a fit for everyday development and Opus 5.5 for more open-ended work that needs sustained judgment.

The useful finish line is a change you can open and check. For a visual change, look at it on a phone as well as a desktop.

For images and video, name the detail that matters

ChatGPT image generation and Nano Banana in Gemini support creating and editing images. Google's current image page lists Nano Banana 2.1 for paid Google AI plans, alongside other available image options.

Bring a reference when the crop, person, object or style needs to stay consistent. Say “keep the person and room; change the background” instead of asking for a whole new image and hoping it keeps the parts you liked.

For short video, Google's current tool is Gemini Omni, including the 1.1 Flash model. It supports generation and conversational editing from different kinds of media. Start with one scene and one clear movement; watch the whole result before building a longer sequence around it.

How to tell whether an upgrade helps you

Choose one task you actually do. Give two available tools the same notes, examples and requirements. Then compare:

Did the important facts survive?
How much editing did you need?
Did you get the file or format you asked for?
Was the wait reasonable for that job?
Could you pick the work back up without explaining everything again?

For background assistants, start with one small, reviewable job and a clear place for the result. If the task includes sending, publishing or changing records, make the approval point part of the brief.

You can keep the tool that works and ignore the next five launch threads. Change when you have a reason: a recurring problem, a missing feature, or a result another tool handles better.

Three prompts you can actually use

For a product description

Turn these notes into a 120-word product description. Keep the measurements, materials and file details exactly as supplied. Give me two opening lines that sound like an actual person. Ask about any missing facts.

For research

Compare what these sources say about this question. Link the page supporting each important claim. Separate the documented facts from your interpretation, and point out anything the sources disagree on.

For a clearer next step

Here is the task and where I stopped. Tell me what is already done, what is missing, and the next three actions in a useful order. Keep the steps small enough that I can start the first one now.

The model name matters less once the job is clear. Start with something you need to finish, give it good material, and judge the result you get back.

Sources and release notes

Checked October 28, 2026.