The best AI apps of October 2026, and what is actually new in each
Ten AI apps worth your time this month, ranked by usefulness rather than hype. For each one: what changed in the last six weeks, who gets it, and where the catch is.
Every big AI app shipped something between mid-September and mid-October. Most of it was model upgrades you will notice without doing anything. Some of it changes what the app is for. This list is the ten apps we would actually recommend this month, and for each one the specific thing that changed, with the date and the plan it needs.
How we picked
Only apps that shipped something verifiable in the last six weeks, checked against the company's own changelog or announcement. Where a feature is still rolling out or in preview, we say so.
1. ChatGPT: GPT-6 for everyone
On 7 October OpenAI moved ChatGPT's paid plans (Plus, Pro, Business and Enterprise) to GPT-6 Sol, and on 8 October the Free and Go plans moved to the lighter GPT-6 Luna. The same release added what OpenAI calls Intelligent UI: answers that mix text with charts, buttons and small forms when the question calls for it. Best for: a general default that most people around you already use. The catch: the free plan gets Luna, not Sol. More on ChatGPT.
2. Gemini: Gemini 4 Argon in the app
Google's new flagship, Gemini 4 Argon, arrived on 13 October in the Gemini app and in AI Mode in Search for Google AI Pro and Ultra subscribers, with Google Antigravity access for Ultra. Best for: people already living in Gmail, Docs and Android. The catch: Argon is not on the free plan. We covered the developer side in the October tooling roundup. More on Gemini.
3. Claude: two new models in three weeks
Anthropic released Claude Sonnet 5.5 on 28 September and Claude Haiku 5.5 on 7 October. Haiku 5.5 is priced at $0.10 per million input tokens for prompts up to 100,000 tokens in the API, which makes it one of the cheapest capable models available. Best for: long writing, careful analysis and coding, especially with Claude Code. The catch: Haiku 5.5's new tokenizer counts about a quarter more tokens for the same text, so the saving is smaller than the headline. More on Claude.
4. Perplexity: research that keeps running
Perplexity's 5 October update added Automations for ongoing tasks, interactive visualizations inside threads, and GPT-6.1 Sol and Claude Opus 5.5 as model options. On 21 September it added an effort setting for its Computer agent and a skills marketplace. Best for: questions that need sources, and research you want re-run on a schedule. The catch: agent runs use compute credits, so heavy use pushes you up a plan. More on Perplexity.
5. Gemini Notebook: still the best way to read your own documents
If you are looking for NotebookLM, it was renamed Gemini Notebook on 16 July; old links redirect. It answers only from the sources you give it, which is exactly what you want for contracts, papers, lecture notes or a stack of PDFs. Best for: students, analysts and anyone who has to read more than they have time for. The catch: it is deliberately narrow; it is not a general assistant. More on it.
6. Cursor: agents you can manage from your phone
Cursor shipped steadily: Projects (10 September, beta), where a coordinator agent plans long work and delegates to other agents; Rollouts and Security Review bots on pull requests (23 September); remote control of agents running on your computer from the Cursor iOS app (6 October); and an app marketplace for its Origin product (13 October). Best for: developers who want an editor built around agents. The catch: remote control only works while your computer is on and online. More on Cursor.
7. GitHub Copilot: Gemini 4 Argon on day one
Gemini 4 Argon became generally available in GitHub Copilot on its launch day, 13 October, for Copilot Pro+, Max, Business and Enterprise, across VS Code, Visual Studio, JetBrains, Xcode, Eclipse, the CLI and the coding agent. Best for: teams that want to switch models without switching tools. The catch: it is rolling out gradually, and organisation plans may need an admin to enable it. More on GitHub Copilot.
8. Microsoft Copilot: Home, Code and Autopilot
On 25 September Microsoft announced a rebuilt Copilot with three parts: Home, where chat and document work meet; Code, which builds small apps, dashboards and automations from a description in a sandbox; and Autopilot, a cloud agent that keeps working on recurring tasks when you are not there. Best for: organisations already on Microsoft 365. The catch: most of it is still rolling out through Microsoft's Frontier programme, and Autopilot is in private preview.
9. Muse from Meta: a free personal agent, US only for now
Meta launched Muse on 8 September: a personal agent that does tasks rather than just answering, running on its own cloud computer, available on iOS, Android, the web and inside WhatsApp. It is free for most needs. Best for: trying an agent with no setup, if you are in the US. The catch: it is rolling out in the US only.
10. Grok: a cheaper frontier model for coders
xAI released Grok 4.7 on 21 September, priced from $2 per million input tokens and $6 per million output in the API, and available in Cursor and xAI's own Grok Build. Best for: developers comparing frontier models on price. The catch: most of the value is in the API and coding tools, not the chat app. More on Grok.
If you only try one thing this month
- Everyday use: stay with the assistant you already have. ChatGPT, Gemini and Claude all improved this month without you doing anything.
- Work from documents: Gemini Notebook.
- Code: try Gemini 4 Argon inside the tool you already use, since Copilot and Cursor both offer it, before you switch editors.
- Agents: Perplexity's Automations are the lowest-effort way to have something run on a schedule.
Kapyn tracks every one of these releases as it happens, in 30 seconds a story. Open the feed or browse the tools directory.
Find these on the Radar
Every tool here lives on Kapyn Radar. Save the ones that fit into a Loadout and find them again.
