AI news October 2026: new models, new prices, new admin controls.

What has changed since late September?

Between 7 and 22 October 2026, OpenAI, Google, Microsoft, GitHub and xAI published new models, prices and administration features. Three kinds of change matter most to a business: new default models in daily use, retiring models in existing tools and new admin controls that decide data access and cost.

This overview groups the announcements by vendor and states plan, rollout and limits as given in the primary source. Prices are quoted as the vendors publish them in US dollars; check taxes, contract terms and your own workspace settings separately before deciding.

GPT-6 and Intelligent UI in ChatGPT chat

On 7 October 2026, OpenAI introduced GPT-6 with Intelligent UI for the Chat experience in ChatGPT. According to OpenAI, responses can include graphics, buttons, forms, charts and interactive elements; simple questions still get text. ChatGPT can also start answering while the model is still thinking.

The rollout started for Plus, Pro, Business and Enterprise, followed the next day by Free and Go. Paid plans use GPT-6 Sol, Free and Go use GPT-6 Luna. In Enterprise, availability depends on admin settings. Important for existing workflows: the models behind Work and Codex do not change with this release.

Our assessment: interactive answers are useful for explanations and quick calculators. A calculator generated in chat is not a reviewed spreadsheet, though; results that feed into quotes or decisions should still be checked.

ChatGPT Business and Enterprise: new controls

Since 20 October, locked chats have been rolling out gradually to all plans. Selected conversations are hidden from history, pins and search and open only with a four-digit PIN or device authentication. Organisational retention and compliance rules still apply, and locked chats remain in the OpenAI Compliance Platform logs.

Since 21 October, owners and admins of Business workspaces can manage members and workspace settings in Admin Console; this does not add Enterprise features. In Enterprise, administrator-managed sync connectors were retired on 16 October; the real-time Google Drive, SharePoint and Microsoft Teams apps replace them. New Sites in workspaces with EU data residency automatically use the existing region for supported storage and hosting.

• Enterprise: review knowledge sources that relied on sync connectors.
• All plans: add locked-chat rules to your internal AI policy.

Codex: GPT-5.5 retired, Ultrafast and new features

Since 14 October 2026 at 19:00 UTC, GPT-5.5 has been retired from Codex for users signed in with ChatGPT. The OpenAI API and Codex sessions authenticated with an API key are not affected. OpenAI recommends GPT-6.1 Sol and explicitly lists workspace defaults, managed configurations, custom agents, scheduled tasks and scripts that still select gpt-5.5.

Since 8 October, GPT-6.1 Sol Ultrafast has been available in Codex and ChatGPT Work on Pro 500 and eligible Enterprise and Edu plans. In Enterprise it is off by default and must be enabled by workspace owners; OpenAI says it supports inference residency in the United States and Europe (EEA and Switzerland). Composer predictions, next-message suggestions in the Codex desktop app, have been in beta since 9 October for personal Pro accounts aged 18 and over only.

OpenAI API: Decisions API, longer caching and voices

The Decisions API has been generally available since 22 October. Built on GPT-6 Luna, it answers bounded questions with a probability, a choice from predefined options or a score on a rubric. Up to 400 questions fit into one request. OpenAI lists $0.10 per million input tokens and $0.01 for cached input; there are no output-token charges. Regional processing is available in the US and EU, with Zero Data Retention for eligible customers.

Other API changes: since 20 October, 12-hour prompt caching can be selected; cache writes cost 2.5 times the input price. Since 21 October, you can set which service tiers each project may use. Since 16 October, prompt-to-voice is available for GPT-Live, along with new voices including the German voices Nuri, Juni and Flitz.

Our assessment: for classification and routing, for example of emails or receipts, the Decisions API is an obvious candidate to test. Use your own examples and define the uncertainty level at which a person decides.

Gemini 4 Argon: Google’s new frontier model

Google made Gemini 4 Argon globally available on 13 October 2026: for developers through the Gemini API (paid tier in AI Studio), for enterprises in Gemini for business, the API and Google Antigravity, and for Google AI Pro and Ultra subscribers in the Gemini app and AI Mode in Search. Google states a unified one-million-token input-output limit.

At launch, an introductory price of $2 per million input tokens and $10 per million output tokens applies, with cached input 95% off. After 30 November 2026, Google says $4 and $20 apply. If you budget now, use the later price.

In GitHub Copilot, Argon has been available since 13 October for Pro+, Max, Business and Enterprise and is billed at provider list prices under usage-based billing. New models are enabled by default unless admins turn them off in the model policy.

Gemini in Google Workspace: skills, Sheets and Meet

Since 13 October, skills in the Gemini app have been rolling out, with completion expected by mid-November. Skills in Workspace are delayed until further notice. A new Customize hub combines skills and connected apps; most connected apps are currently available in English in the US only. Admins manage integrations in the Admin console under Apps > Additional Google services > Gemini.

Since 19 October, Gemini in Google Sheets can analyse BigQuery data via Connected Sheets in natural language and build dashboards with pivot tables and charts. It is available on Business Standard and Plus and Enterprise Standard and Plus, among others; it is on by default with no admin control. Since 22 October, Google Meet translates speech between 63 languages, for up to 90 minutes per meeting on base Workspace plans; recordings and client-side encrypted meetings are excluded.

Our assessment: because the BigQuery analysis ships without an admin control, existing BigQuery permissions alone decide who can query which data. Review those permissions before the rollout.

Microsoft 365 Copilot: Office in the Copilot app

Since 9 October, Microsoft has been bringing Word, Excel and PowerPoint into the new Copilot app, initially for Frontier customers in Chat and Cowork. Documents open next to the conversation, support co-authoring and are stored in OneDrive; according to Microsoft, formatting, sharing permissions, version history and sensitivity labels are retained.

The 6 October release notes also list chat directly in Microsoft 365 Copilot Search, a regenerate action including model switching and admin-editable query string and identity mapping rules for Copilot connectors. After an admin changes a connector rule, they should check with test users that search results are still shown only to authorised people.

GitHub Copilot: models deprecated since 19 October

Since 19 October 2026, several models have been deprecated across most Copilot experiences. GitHub lists GPT-6.1 Sol as the alternative to GPT-5.5 and GPT-5.4, GPT-6 Luna for GPT-5.4 mini and GPT-5 mini, Gemini 3.8 Flash for Gemini 3.7 Flash and Grok 4.7 for Grok 4.5. Enterprise admins may need to enable alternative models in the model policy.

• Find and replace saved model names in scripts, agents and team defaults.
• Decide whether new models such as Argon should be enabled automatically.

xAI Grok: Voice 3.0, X Subscription and partner network

According to the xAI release notes, Grok Voice 3.0 has been available in the Speech-to-Speech API since 14 October at the same price as version 2.0: $0.08 per minute. The grok-voice-latest alias was switched automatically; grok-voice-2.0 is deprecated. If you run voice agents pinned to a model name, test before switching.

With The X Subscription, xAI bundles Grok, Grok Bot, Cursor and X Premium in one plan: Free, Plus at $8, Premium at $30, Super at $100 and Ultra at $200 per month. Teams get Premium, Super and Ultra at the same per-seat prices, including SSO, admin controls and SOC 2 compliance; existing SuperGrok, Cursor and X Premium plans can stay as they are. On 22 October, xAI also launched a partner network for transformation and marketplace partners.

What should a small business do now?

Start with changes that affect current work: replace saved model names in Codex and GitHub Copilot, replace the retired sync connectors in ChatGPT Enterprise and review permissions before BigQuery analysis is used in Sheets. Optional tests follow, for example with the Decisions API, Gemini 4 Argon or Office in Copilot, each on a clearly bounded task.

Do we need to switch immediately because of new models? No. Compare real tasks, including expert review. Do these features apply in Germany? A global announcement, an EU data region and enablement in your own workspace are three different claims. The free AI System Check helps organise prerequisites and narrow down a useful first application. The examples here are suggested tests, not customer results.