Today's AI news: OpenAI models on Google Cloud, DeepSeek's open-source inference milestone, ChatGPT Ads in 26 new countries, Mistral Large 4, and more.
Here is everything that mattered in AI today, Wednesday 21 October 2026.
OpenAI Models Are Coming to Google Cloud
In one of the more eyebrow-raising partnership announcements of the year, OpenAI and Google Cloud have confirmed that OpenAI's frontier models — along with Codex workflows — will be available through Google Cloud's Gemini Model Garden and the Google Cloud Marketplace. The move means enterprise developers can access OpenAI's technology directly within Google's cloud infrastructure, a significant commercial détente between two of the industry's fiercest rivals. No pricing details were included in today's announcement, but the availability of Codex workflows suggests a strong focus on software development use cases.
DeepSeek Drops an Open-Source Inference "Bomb"
Chinese AI lab DeepSeek has released what analysts at Latent Space are calling an open-source inference milestone. DeepSeek DeepInfer is reported to be running 155 trillion tokens per day per cluster, generating an estimated US$22 million per day at roughly 80% margin — implying a cluster payback period of just 81 days. The sheer throughput and economics, if accurate, represent a striking challenge to the assumption that frontier-scale inference requires proprietary, closed infrastructure. The open-source release means any well-resourced team can, in principle, study or deploy the same approach.
ChatGPT Ads Expands to 26 New Countries
OpenAI has confirmed that its ChatGPT Ads platform is expanding to 26 additional countries, with the company reporting that more than 100,000 advertisers have now used the platform. The expansion targets users who are actively exploring, comparing, and deciding what to buy inside ChatGPT — positioning the product as a commercial-intent channel rather than a pure information tool. The milestone signals that OpenAI is accelerating its advertising revenue strategy at meaningful scale.
Mistral Releases Mistral Large 4
French AI company Mistral has published documentation for Mistral Large 4, its latest flagship model, drawing significant attention on Hacker News. While detailed benchmarks were not included in today's linked material, the release continues Mistral's pattern of competing at the frontier of large language model capability, particularly for European and enterprise audiences. The model's documentation is now live on the Mistral site.
AT&T Uses OpenAI to Rethink In-House Legal Work
OpenAI has published a case study detailing how AT&T's internal LegalEdge team is using OpenAI models to make legal knowledge more accessible across the organisation. The project focuses on rethinking intake processes and scaling legal judgement — essentially allowing non-specialist staff to access structured legal guidance more efficiently. It is a notable enterprise example of AI being applied not to replace lawyers but to extend the reach of in-house legal teams across a large corporation.
Google Staff Resign Over Military AI Contracts
A number of Google employees have resigned in protest over the company's AI contracts with the military, according to a BBC News report circulating on Hacker News today. The walkout echoes earlier waves of employee activism at Google — most notably the 2018 Project Maven controversy — and signals that internal dissent over the ethics of AI in defence applications remains a live issue at major technology companies. The specific contracts at the centre of today's resignations were not detailed in the available material.
AI Agents Used to Uncover Leads in Brain Cell Data
OpenAI, in collaboration with PerturbAI and Scripps Research, has published details of a project that deployed thousands of distributed AI agents to analyse data from millions of brain cells. The goal was to surface promising research leads in complex biological datasets that would be impractical to review manually. The work is an early illustration of how large-scale agent orchestration — rather than a single model — is being applied to accelerate scientific discovery in life sciences.
Frequently Asked Questions
Are OpenAI models now available on Google Cloud?
Yes. As of 21 October 2026, OpenAI and Google Cloud have announced that OpenAI's frontier models and Codex workflows will be available through Google Cloud's Gemini Model Garden and the Google Cloud Marketplace.
What is DeepSeek DeepInfer and why is it significant?
DeepSeek DeepInfer is an open-source inference system from Chinese AI lab DeepSeek. It is reported to run 155 trillion tokens per day per cluster, generating roughly US$22 million per day at about 80% margin, with an 81-day cluster payback period. Its open-source release challenges the assumption that frontier-scale inference requires closed, proprietary infrastructure.
How many countries does ChatGPT Ads now operate in?
Following an expansion announced on 21 October 2026, ChatGPT Ads is now available in 26 additional countries. OpenAI also reported that more than 100,000 advertisers have used the platform.
Why are Google employees resigning over AI contracts?
According to a BBC News report published on 21 October 2026, a number of Google staff have resigned in protest over the company's AI contracts with the military. The specific contracts were not fully detailed in available reporting, but the situation echoes earlier employee activism at Google over AI use in defence applications.