AI This Month Review
· October 2026

THE MONTH IN ONE LINE
This month the chat window became the place to make the file itself: documents, slides and spreadsheets are now built and edited inside ChatGPT, Claude and Gemini.
By The Numbers
The stats worth knowing this month, distilled from the data behind the headlines.
94%
Less cash reconciliation time
An independent Schlotzsky's and Cinnabon franchisee cut cash reconciliation time by 94% and menu repricing time by 95% using ChatGPT Work - the kind of back-office chore every small business has.
10-15 hours
Weekly admin time freed at The Den
The Den Family Social's small leadership team saves 10-15 hours a week with ChatGPT Work, preparing grant applications in 2 hours instead of 3 days while opening a second location.
954%
Credit Saison pilot ROI
Credit Saison reported a 954% ROI from a ChatGPT Enterprise pilot, saving 14 hours per participant a month - a result that came with strong leadership buy-in and a dedicated AI team.
75% to 56%
Enterprises trusting unsupervised agents
In a VentureBeat survey of its readers, the share of enterprises letting AI agents make production changes unsupervised fell from 75% to 56% between July and August 2026 - a reminder to keep a human check on agent work.
More than 5x
Gartner's agent cost forecast
Gartner expects inference costs (the cost of running the model) per agentic workflow (a multi-step job an AI agent carries out) to rise more than fivefold through 2028 even as token prices fall, because agents use 5-30 times more tokens per task than basic chatbots.
76x
Asana browser-agent cost cut
Asana cut model costs for its browser agent 76x and made it 5x faster in tests, with an optimised run averaging $0.47 - the saving came from refactoring and caching work, not just a cheaper model.
$368B
Southeast Asia digital economy 2026
The e-Conomy SEA report projects Southeast Asia's digital economy at $368 billion in 2026, with generative AI a daily tool for 243 million people in the region.
The Core Shifts
The structural changes shaping how AI tools work, cost, and get used - drawn from patterns across the month's coverage.
Assistants make the finished file
In October the big assistants stopped handing back text to paste and started producing the file itself. The Gemini app now generates, previews and edits documents, PDFs, presentations and spreadsheets inside the chat, with a full-screen editor for direct changes, and it is available to all Gemini users. OpenAI's GPT-6 brought Intelligent UI to ChatGPT, so an answer can arrive as a diagram, a side-by-side comparison or a small built-in tool such as a calculator, and it is rolling out to Free and Go users as well as paid plans. Anthropic added Claude Dashboards, which builds live, auto-updating dashboards from connected data, and Claude Motion, which turns a prompt into an animated explainer. ChatGPT and Google Slides also gained deck building and whole-deck editing.
The honest limit is unevenness. In Gemini, PDF preview and editing work on web and mobile, but the other formats are web-only for now. Intelligent UI does not work with ChatGPT's Pro reasoning option, and Free and Go users get GPT-6 Luna while paid plans get GPT-6 Sol. Dashboards is a beta on paid Claude plans and Motion a beta on Team and Enterprise only; Motion generates animation as code, so it suits text, charts and shapes rather than video footage. Whatever the tool, the file is a first draft you still need to check.
Free Access: The Gemini app now generates, previews and edits documents, PDFs, presentations and spreadsheets in the chat, for all users.
What this means for you
Next time you would open a blank document or spreadsheet, ask the Gemini app to draft it in the chat, then fix it in the full-screen editor (on the web) before you export it.
Ads and shopping enter AI answers
AI answers became a place to sell. OpenAI expanded ChatGPT Ads to 26 more countries, 89 in total, and says over 100,000 advertisers have used the platform; it also began testing visual ads with images, shown during image generation for US Free and Go users. Its engineering write-up says the ad system is kept separate from ChatGPT's answers. Google took sponsored results in AI Mode to 10 new countries, including Australia, Canada and India. Shopping moved in too: Google's Universal Cart in the US saves products from several retailers in one place in Search and the Gemini app, tracks price drops and finds promo codes, and Albertsons added a Safeway plugin to ChatGPT that builds a grocery cart from recipes, with checkout on Safeway's site.
The limits sit with who sees ads and how clearly. ChatGPT ads appear only on the Free and Go plans; Plus, Pro and Enterprise stay ad-free. Google labels AI Mode ads 'Sponsored', but inside an AI-written answer they may blend in more than classic search ads. Universal Cart is US-only and depends on merchants joining Google's commerce protocol, so it is still worth comparing prices yourself.
Adoption: ChatGPT Ads now runs in 89 countries, and over 100,000 advertisers have used the platform.
What this means for you
If you sell something, look at ChatGPT's self-service Ads Manager and AI Mode placements in your Google Ads account as new channels. As a user, read any 'Sponsored' item inside an AI answer exactly as you would an ad.
Voice moves into everyday apps
September's review covered real-time voice arriving mostly as developer APIs; this month it landed inside apps people already use. ChatGPT now accepts audio uploads on paid plans, turning meeting, interview or lecture recordings into transcripts, summaries and answers to questions. ChatGPT Voice moved to GPT-Live-1.1, with more natural accents; its lighter Quip version is available, with limits, even on the Free plan. Google Meet's live speech translation now covers 63 languages and detects them automatically, even with several languages in one meeting. Docs Live on mobile lets you draft a structured Google Doc by talking, pulling context from your emails, chats and Drive files. Guided Vision in Gemini Live, built with the blind and low-vision community, describes what your phone camera sees and gives spoken cues.
The catches are access and accuracy. Audio uploads are not on the Free plan, and transcription quality depends on audio clarity, accents and background noise. Meet translation runs up to 90 minutes per meeting on standard eligible plans and does not work in recordings, live streams or dial-in calls. Docs Live is English-only and limited to paid Workspace and Google AI plans. Google says Guided Vision can make mistakes and is not a replacement for a cane or other mobility aid.
Language Reach: Google Meet's live speech translation now covers 63 languages and detects them automatically.
What this means for you
If you are on a paid ChatGPT plan, upload your next meeting recording and ask for action items and a draft follow-up email, then check the key points against the recording before sending.
Frontier ability for less
September's review covered cheaper models claiming flagship-level ability; in October the cheap tier grew again and routing between models joined in. At the top, Google made Gemini 4 Argon its leading model for Google AI Pro and Ultra subscribers, Meta launched Muse Wonder for agents at $4 per million input and $20 per million output tokens (tokens are the chunks of text models are billed by), and Moonshot's Kimi K3.1 Pro arrived at $2 and $10. Below them, Anthropic's Claude Haiku 5.5 and OpenAI's GPT-6.1 Terra targeted routine work at lower cost, and MiniMax's M3.1 Flash, at $0.15 and $0.60, beat DeepSeek and Claude Opus 5 on a key business benchmark. GitHub's Copilot Fusion takes another route: it sends each coding task to a suitable model, and in independent testing its Max level came close to Claude Opus 5.5 at about half the per-task cost.
Cheaper is not the same as reliable. When M3.1 Flash gets an answer wrong, it hallucinates 93% of the time rather than admitting uncertainty. Fusion's savings are per task and leave out human review, retries and security work, results are uneven across jobs, and the High and Max levels need Pro+ at $39 a month or above. Kimi K3.1 Pro faces unconfirmed allegations that influencers were paid to promote it, so treat its launch praise with care.
Value Option: In independent testing, GitHub Copilot Fusion Max came close to Claude Opus 5.5 at about half the per-task cost.
What this means for you
Before paying for the top model, run one routine task you do every week on a cheaper one you can already reach, such as GPT-6.1 Terra in ChatGPT Work on Plus, Pro, Enterprise or Edu, and compare the results side by side.
Image AI learns to edit
July's review covered image generation becoming a normal button in ordinary tools; this month the tools turned to editing what you already have. Canva Brush edits existing designs from a text prompt, from restyling colours to sweeping changes across multi-page documents, and Canva reports it is up to 69% faster and 46% cheaper per edit than previous design agents. Microsoft's MAI Image 2.7 edits images rather than creating them: it restyles or relights a scene, changes text inside an image, restores old assets or moves a product into a new setting, and it debuted third on Arena's image-editing leaderboard. Magnific One puts an AI creative director between your prompt and the image, applying a company's brand kit and flagging off-brand results. Canva's new Craft model also splits generated images into editable layers.
Most of this is early or gated. Brush is a beta for Canva Pro subscribers, and each edit counts against the monthly AI allowance. MAI Image 2.7 is a preview that only edits for now. Magnific One runs on OpenAI's GPT Image 2, starts at $14.50 a month billed annually, and its quality claims come from its own blind comparisons, not independent tests. Leaderboard positions are launch-day snapshots and can change.
Efficiency Claim: Canva reports Brush is up to 69% faster and 46% cheaper per edit than previous design agents.
What this means for you
Take one graphic you reuse, such as a flyer or a social post, and ask an editing tool to restyle or retext it instead of generating a new one; MAI Image 2.7 can be tried in MAI Playground.
Telling AI from real
Tools for spotting AI-made content moved closer to ordinary users. Google's SynthID Detector is now available globally in English and free: you upload an image, video or audio file and it checks for the invisible watermark used by Google's AI and partners including OpenAI, NVIDIA and Kakao. OpenAI said that, over the coming weeks, text EU users get from ChatGPT and Codex will carry a hidden watermark in its word choices that survives copy-paste and does not identify the user, to comply with the EU AI Act. OpenAI also said it will honour a common robots.txt signal so website owners can state their AI-training preferences once for all participating providers.
Detection remains partial. SynthID only recognises watermarked content from Google and its partners, so a clean result does not mean a file is real. OpenAI's text watermark is easy to weaken: in its own tests, swapping 10% of words for synonyms cut detection from about 92% to 66%, and its detector is open only to approved researchers and organisations. The robots.txt signal is voluntary and is not active yet.
Detection Limits: In OpenAI's tests, swapping 10% of words for synonyms cut text-watermark detection from about 92% to 66%.
What this means for you
Next time an image or clip looks too good to be true, run it through the free SynthID Detector, and read a 'no watermark' result as 'unknown', not as proof it is real.
If you could only do 4 things...
Concrete, low-effort moves you can make today - no research required.
What to do:
“Next time a text message looks suspicious, long-press the home button and circle it with Circle to Search to check it for scam signs before you tap any link.”
'Open' does not mean free for your business
October brought a wave of open-weight models (model files you can download and run yourself), and most came with strings. FLUX.3 Image Klein is free only for individuals and businesses under $5M in annual revenue; Liquid AI's LFM2.5 and LFM3 allow free commercial use below $10M; companies with more than $20M in monthly revenue need a commercial licence for Mistral's Timestral. Meta's Muse Spark Open has limited commercial terms, requires a 'Built with Muse Spark Open' credit and needs approval to download. Mistral Large 4's licence requires a separate commercial licence for anyone selling it as a model service, and NVIDIA's Nemotron weights are free, but production use of NVIDIA's NIM microservices needs a $4,500 per GPU per year licence. Running an open model also moves safety and compliance onto you, since its guardrails can be easily reversed.
What This Means For You
Before you build a product or client service on an open model, read its licence for revenue limits, attribution rules and service restrictions - 'open weights' only means you can download it.
Canva Craft
Canva's first in-house image model, trained on human-designed templates such as posters, social posts, menus and infographics rather than photos. It plans the layout and text before drawing, so on-image text is readable, in 55 languages, at up to 2K resolution and about 30 aspect ratios. Magic Layers splits results into editable layers, and Craft debuted fourth on the DesignArena image leaderboard. It is free to try in Dream Lab with any Canva account, including free ones; monthly AI use is capped by plan.
Best For
Small business owners, café and shop owners making menus or posters, and social media managers who need the same post in several sizes.
Skip If
Heavy users on a free plan who will hit the monthly AI cap, and team or school accounts where an admin has switched Dream Lab off.
Practical Action
Make this week's specials poster or menu board in Dream Lab, then fix any wording in the editable layers before printing.
Try This Prompt
Paste into Canva Dream Lab: 'A weekly specials poster for a small café with three dishes and prices: [dish - price], [dish - price], [dish - price]. Warm, hand-drawn style.'
Head-to-Head
When multiple vendors ship the same capability, here's how your options compare.
Make slides and documents inside your AI assistant
Three vendors shipped ways to build or edit presentations and documents without leaving the assistant; which one you already have depends on your plan.
Create, edit and present slides inside ChatGPT, collaborate with teammates in real time, and export to PowerPoint, PDF or Google Slides. Rolling out to Pro (from $100/month), Business and Enterprise accounts with Space access.
All three are out of beta and available on every Claude plan, including Free, with shared editing of the same document, deck or design.
Edit an entire existing deck from one prompt - update content throughout, refresh the style or match another presentation - as native, editable slides. English only; needs Workspace Business or Enterprise Standard/Plus, or Google AI Pro or Ultra. The free Gemini app also drafts presentations in chat.
What This Means For You
Already on ChatGPT Pro or Business, a Workspace Business or Enterprise Standard/Plus plan, or Google AI Pro or Ultra - you have one of these included. On a free plan, start with Claude's Docs and Slides or a draft in the Gemini app.
What Changed In Tools You Already Use
Updates and new features rolled out this month to the tools already in your stack.
ChatGPT
OpenAIGPT-6 with Intelligent UI (rolling out to Free and Go); audio uploads (paid plans); GPT-Live-1.1 and new voices; ChatGPT Slides (Pro, Business, Enterprise); GPT-6 Pro update with parallel agents (Pro); GPT-6.1 Terra in Work and Codex (Plus, Pro, Enterprise, Edu); Texting your dot beta (US Pro); monthly API credits for Pro and Business; more Work and Codex usage on Pro and Business Premium; account switching on iOS and Android; teen flashcards and quiz creation; stronger teen protections (rolling out); text watermarking in the EU; visual ad test and ads in 26 more countries (Free and Go); ServiceNow, Atlassian and Safeway (US grocery) plugins.
Key Impacts
- ›Answers can now arrive as diagrams, comparisons and small tools, on free plans too.
- ›Paid users can upload recordings for transcripts and summaries.
- ›On Free or Go, expect ads; Plus and above stay ad-free.
Claude
AnthropicHaiku 5.5 model launched; Docs, Slides and Design out of beta on all plans; Dashboards (beta, paid plans) and Motion (beta, Team and Enterprise); Claude Duet proactive agent (research preview, waitlist for US Pro and Max); chat and Cowork unified for Enterprise; Enterprise smart reports, surveys and pooled budgets; Claude for Nonprofits plugin; legal artifacts in Claude for Legal (Enterprise); Claude Code mods; Code Review generally available (Team, Enterprise); Claude Security continuous scanning (Enterprise beta).
Key Impacts
- ›Free users can now make docs, slides and designs in Claude.
- ›US Pro and Max users can join the Duet waitlist; it shares your usage limits, and background work uses them faster.
Gemini app
GoogleGemini 4 Argon (Google AI Pro and Ultra); create, preview and edit documents, PDFs, slides and spreadsheets in chat; publish apps to Google Sites; new customize hub; skills rolling out, replacing Gems; Personal Intelligence in Android Auto; Guided Vision in Gemini Live (Android); small-business tools; student hub for Education accounts and immersive explorations; Gemini Notebook as a connected app; Universal Cart (US); Chrome autofill from Gmail (US).
Key Impacts
- ›Anyone can draft and edit files inside the chat (most formats web-only for now).
- ›Skills replace Gems - check how yours carry over.
- ›Pro subscribers get Argon at no extra cost.
Gemini in Workspace
GoogleWhole-deck edits in Slides; Gemini in Sheets on iOS; plain-language BigQuery analysis in Sheets; Fill with Gemini adds cell-level web citations and 3,000-cell requests; Gmail AI Inbox (US, paid editions); Meet next steps and meeting recaps in Chat; Meet speech translation in 63 languages; Docs Live voice drafting on mobile (English); Vault exports of Gemini conversations.
Key Impacts
- ›Fill with Gemini now shows the web source for each cell - check it before relying on the data.
- ›US paid Workspace users get AI Inbox to-dos and thread summaries.
Grok
xAIThe X Subscription bundles Grok, Grok Bot, Cursor and X Premium in Free, Plus, Premium, Super and Ultra tiers.
Key Impacts
- ›One plan now covers Grok chat and Cursor coding with a shared usage pool.
- ›Third-party models in Cursor are billed separately at API list prices on Premium and above.
Gemini Notebook
GoogleVideo overviews, interactive learning overviews and more quiz formats; guided slide decks and video overviews with a review step; hundreds of free textbooks (US and India providers); teacher insight summaries in Google Classroom.
Key Impacts
- ›You can now review and steer a deck's outline before it renders.
- ›Students can ground answers in free textbooks, with citations.
Perplexity Computer
PerplexityProjects, shared workspaces for people and agents (Pro, Max); a fully local version on Apple silicon Macs with model, harness and tasks on-device (Pro, Max, Enterprise); 10 ready-made finance and planning Skills for US Amex Business Card Members (Enterprise plans).
Key Impacts
- ›Private files can stay on your Mac, but 36 GB of RAM or more is recommended.
Canva
CanvaCraft, Canva's own image model, in Dream Lab for all users and via its API; Brush, prompt-based editing of existing designs (beta, Canva Pro).
Key Impacts
- ›Free accounts can try Craft, within the monthly AI cap.
- ›Pro users can restyle existing designs with a prompt.
Theme Spotlight
A closer look at the concepts and ideas worth understanding this month.
How AI looks things up before it answers
RAG is how an AI tool looks things up before answering: it pulls relevant material from the web, a company knowledge base or your own files and hands it to the model along with your question. That reduces made-up and outdated answers, and it is why some tools can point to their sources. It works alongside the model rather than replacing it. To go further, Perplexity's guides describe ten RAG architecture types and include a beginner's Python tutorial.
Where You Meet It
AI web search, support chatbots such as Zendesk's, and Gemini reading your calendar all use it.
The Trade-off
It adds setup complexity, and long-context models can sometimes fit all the material into the prompt instead.
Try It
Next time a vendor pitches you a support chatbot, ask whether it answers from your own help docs and what it does when it is unsure.
Why the same model behaves differently in different tools
The harness is everything around an AI model except the model itself: how it manages context, which tools it can call, its memory, its permissions and how it checks its work. It is what lets an assistant remember a conversation, look things up, run code and finish multi-step tasks.
Cost Finding
A Shandong University study found a coding agent's harness can triple its cost without making it more accurate.
What You Control
Standing instructions, connected apps and approval rules are the parts of the harness users can shape.
Try It
Set one standing instruction in your AI tool, such as your preferred format, and watch whether it applies across your next few tasks.
AI that builds a world you can steer
A world model generates an interactive simulation in real time that you can steer with plain language - for example changing the weather or adding characters to a scene. Google DeepMind's Genie 4 adds native audio, generating ambient sound, music and dialogue as the scene plays, and draws on the Gemini family to keep scenes realistic.
Availability
Rolling out to trusted testers; coming soon to Google AI Ultra subscribers in Project Genie (US only, 18+).
Named Uses
Pre-visualising film scenes and prototyping game environments.
Price & Access Tracker
Open weights released on Hugging Face under the Mistral Ecosystem Open-Weight licence; API use (the paid connection developers use to plug a model into their own apps) is $0.70 per million input tokens and $2.10 per million output tokens.
API at $1 per million input and $5 per million output tokens, 50% below GPT-5.6 Terra and GPT-6.1 Sol; included for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex.
$0.10 per million input and $0.50 per million output tokens for prompts up to 100,000 tokens; Anthropic cites a 75% lower average running cost than its predecessor.
Pro 100, 200 and 500 subscribers can claim $100, $200 or $500 a month in API credits; Business gets $20 per Standard seat and $100 per Premium seat, capped at $500 a month. Credits expire at the end of each billing cycle.
Allowance doubled on Pro 100 and Pro 500, previous allowance restored on Pro 200, and more usage on Business Premium.
Included for verified nonprofits on Team and Enterprise; nonprofit Team seats are $8 per user a month, or $3 in low- and middle-income countries.
One plan in five tiers: Free ($0) with daily Grok access and a Cursor starter allowance, Plus ($8/month), Premium ($30), Super ($100) and Ultra ($200).
The new streaming transcription model costs $0.54 per hour of audio at an introductory price through year-end; MAI-Voice-2.1 is $22 and MAI-Voice-2.1-Flash $15 per million characters.
Until March 29, 2027, enrolled cards get a one-time $125 statement credit on a qualifying Enterprise purchase of $325 or more, plus 1,000 bonus Computer Credits.
What This Means For You
If you pay for ChatGPT Pro or Business and build with the API, claim this month's API credits before your billing cycle ends - unused credits expire.
AI Milestones & Watchlist
Things To Try This Weekend
Hands-on projects you can knock out in a weekend with the tools you already have.
Build a small interactive tool, such as an FAQ helper or a quiz, and publish it as a web page you can share.
Standard Tool
Gemini app and Google Sites
Requirements and Cost
Included with the Gemini app for personal Google accounts, Workspace customers and Workspace Individual subscribers; you need permission to create Google Sites in your organisation.
- 1Open the Gemini app and describe the small interactive tool you want, who will use it and what it should do.
- 2Test it in the chat and ask Gemini to adjust the content and behaviour until it works the way you want.
- 3Publish the app to a new private Google Sites draft.
- 4Customise the page in Google Sites, then share it with your team, customers or students.
Turn your notes into a slide deck you have steered, not just generated.
Standard Tool
Gemini Notebook
Requirements and Cost
Free with a Google account at standard limits; Google AI Plus, Pro ($19.99/month) and Ultra raise limits.
- 1Add your notes or source documents to a notebook in Gemini Notebook.
- 2Ask for a slide deck and choose from the focus topics, styles and formats it suggests.
- 3Turn on the review step, read the outline and ask for changes, such as adding an executive summary or reordering sections.
- 4Generate the deck and check every fact against your sources.