October 2026 AI updates: what changed for businesses

October 2026 brought two new frontier models from Google, a faster computer-use model from OpenAI and a much cheaper small model from Anthropic. Assistants that keep working in the background also moved closer to everyday use. Here are the October 2026 AI updates that matter for a business, and what each one means.

New frontier models from Google and OpenAI

Gemini 4 Argon started rolling out globally on 13 October. Developers and businesses can use it through the Gemini API and Google Cloud. Google AI Pro and Ultra subscribers can use it in the Gemini app and AI Mode in Search. The API costs US$2 per million input tokens and US$10 per million output tokens until 30 November 2026. After that, the price is US$4 and US$20.

Gemini 4 Carbon followed on 28 October. Google calls it its strongest model yet for long-running engineering work and cybersecurity defence. For now, it goes to a subset of cyber defenders only. Developers, businesses and consumers will get it “as soon as possible”.

GPT-6.1 Terra arrived on 26 October, offering “fast and capable computer use” for everyone. Computer use means the model operates websites and software on screen.

On 7 October, OpenAI began rolling out GPT-6 in ChatGPT globally. Answers can now include visuals and interactive elements, known as Intelligent UI.

For the full GPT-6 family, read GPT-6 explained. For Google’s lineup, read Gemini 4 Argon and Carbon explained.

Small models got cheaper

Small models handle high-volume work such as sorting emails, tagging data and drafting short replies. Their prices fell again this month.

• Claude Haiku 5.5 (7 October) costs US$0.10 per million input tokens and US$0.50 per million output tokens, for prompts up to 100,000 tokens. Anthropic says it costs around 75% less to run than Haiku 4.5 on average.
• Mistral Large 4 entered public preview on 6 October. Its open weights followed on 29 October, so businesses can run it on their own servers.
• Meta released its Muse Wonder models on 12 October through the Meta Model API.

Cheaper models change the maths on automation. A task that was too costly to run thousands of times a month may now be affordable.

Assistants that work in the background

Several releases this month moved AI from answering questions to carrying out work.

• Gemini agent (8 October) is Google Cloud’s agent for work. It runs long tasks in the cloud and can pick between Gemini and Claude models for each job. Businesses can set hard spending caps per project.
• The Gemini app gained a tasks mode on 13 October, and now connects to more than 40 apps. Tasks mode lets Gemini keep working when you are away from your computer.
• Claude Duet (21 October) is a research preview of a proactive agent app. It checks with you before it acts on your behalf. Access is by waitlist, starting with Pro, Max and Ultra subscribers in the US.
• Microsoft added usage-based billing for Work IQ APIs in late October. Developers can now call Copilot agents and models without pre-assigned licences.

To see what this looks like in practice, read 7 AI automations a small business can run today.

Search, ads and your website

Four changes this month affect how businesses are found online.

• More links in AI answers. On 15 October, Google announced it is showing inline links higher up in AI Mode and AI Overviews more often. It is also testing a section for official websites at the top of some answers.
• Ads in AI Mode expanded to 10 more countries on 28 October, including Australia, India, Malaysia, the Philippines and Singapore.
• ChatGPT Ads expanded to 26 more countries on 21 October. More than 100,000 advertisers have used the platform.
• AI training controls became clearer on 30 October. Google and OpenAI both said they will support a shared robots.txt signal for opting out of AI training, based on IETF work. Google aims to support it in 2027. Bing introduced separate settings for training and for grounding, which take effect on 1 February 2027.

The training and grounding settings do different jobs. Blocking training keeps your content out of future models. Blocking grounding can keep your pages out of live AI answers. Bing warns that blocking grounding reduces brand visibility in AI experiences. Most businesses want to appear in those answers. Read more in Google AI Mode in 2026.

Voice assistants in more languages

• Siri AI became available in French, Japanese, Korean, Portuguese and Spanish on 26 October. Apple plans Chinese (Traditional), Norwegian, Turkish and Vietnamese by the end of the year.
• Grok Voice 3.0 from xAI launched on 14 October as a speech-to-speech model at US$0.08 per minute.

Models on more clouds, and Anthropic in Singapore

Claude models became generally available on Google Cloud, managed by Anthropic, on 22 October. On 21 October, OpenAI announced its models are coming to Google Cloud’s Model Garden.

Anthropic also opened an office in Singapore on 14 October, its fifth in Asia-Pacific. It signed an agreement with Singapore’s Ministry of Digital Development and Information, which includes helping small and medium-sized businesses adopt AI responsibly.

What the October AI updates mean for your business

• Build so you can switch models. The best model for a task changed more than once this year. Workflows that can swap models keep improving as prices fall.
• Revisit tasks you ruled out on cost. Small models now handle routine work for much less than a year ago.
• Check your website’s AI settings. Make sure you are not blocking the AI search crawlers you want to be found by.
• Budget for price changes. If you use Gemini 4 Argon through the API, its price doubles after 30 November.