What's Changed in AI This Week? OpenAI GPT-6.1 Terra, Microsoft Copilot, Gemini 4 Carbon and Anthropic's IPO
Cheaper computer automation, more choice in Microsoft Copilot, new business integrations and the growing costs behind AI.
AI is getting cheaper to automate with, easier to connect to business software and more flexible in how it's deployed. This week, OpenAI introduced GPT-6.1 Terra, Microsoft opened up its Copilot harness, and Oracle announced new ChatGPT integrations. Google is also pushing AI further into cybersecurity, while Anthropic's IPO filing gives us a clearer picture of the costs behind the technology.
AI continues to move frighteningly* quickly, with another busy week of announcements from some of the biggest names in technology, so let's unpack what's new this week. (*That's the only Halloween reference we'll make this week. Honest.)
OpenAI has introduced GPT-6.1 Terra, making computer-based automation more affordable, while Microsoft is giving businesses more flexibility over the AI models they use through Copilot. Oracle is bringing ChatGPT closer to everyday business applications, and Google's latest Gemini model is pushing further into cybersecurity.
We've also had a closer look at the costs behind AI, with Anthropic's IPO filing revealing just how much money is being spent on infrastructure to support its growth.
For businesses, there's plenty to consider. AI tools are becoming more capable and accessible, but giving them access to company systems and data means getting security, permissions and costs right from the start.
This week's AI headlines
OpenAI launches GPT-6.1 Terra
A faster model for visual reasoning and computer use, with lower token costs that could make everyday automation more affordable.
Microsoft opens up its Copilot harness
Businesses and developers can use a managed agent framework across more than 20 AI models through Microsoft Foundry.
OpenAI and Oracle expand their partnership
New integrations aim to connect ChatGPT with finance, procurement, planning and other approved business workflows.
Google introduces Gemini 4 Carbon
Google's latest model focuses on long-running tasks and cybersecurity, with initial access limited to selected cyber defenders.
Anthropic publishes its IPO filing
The filing reveals rapid revenue growth, substantial infrastructure spending and the importance of usage-based AI pricing.
OpenAI's GPT-6.1 Terra makes computer use cheaper
OpenAI introduced GPT-6.1 Terra, a faster model aimed at the visual and computer-based work that fills an ordinary working day. OpenAI's examples include reading invoices and entering details into accounting software, filling in forms, interpreting technical drawings and editing video.
The price is the part most businesses should notice. Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. OpenAI says that's 50% below GPT-5.6 Terra and GPT-6.1 Sol, while its own tests show a substantial improvement in visual reasoning, computer use and factual accuracy.
Those are vendor-reported benchmarks, so they're worth testing against real business tasks before making purchasing decisions.
Terra is available through the API and is rolling out in ChatGPT Work and Codex. The ChatGPT release notes list Plus, Pro, Business, Enterprise and Edu availability, subject to region and workspace settings. OpenAI also says the same Work and Codex allowance should cover more Terra tasks than work performed with its larger models.
Computer-use models can operate existing software without waiting for every application to expose a perfect API. That could make automating repetitive admin much easier, especially for smaller businesses using a mixture of systems.
There are risks to consider, though. An AI agent that can click, type and submit information can also make mistakes or take actions it shouldn't.
Starting with low-risk tasks in a test environment, keeping human approval in place for financial or customer-facing actions and measuring the cost per completed task will help businesses understand whether the technology delivers value.
Microsoft opens up the technology behind Copilot
Microsoft has made its Copilot harness available through Microsoft Foundry.
In normal person speak, that means the harness is the machinery around an AI model. It brings together instructions, context, tools, an execution environment and the controls that allow an agent to complete longer pieces of work.
The same layer already powers Copilot Code, Copilot Studio, the GitHub Copilot app, the command line and Visual Studio Code.
Microsoft says it works across more than 20 frontier models, supports bring-your-own-key arrangements and can route work to the appropriate model. Teams can download the software development kit (SDK) or run the harness as a managed cloud service in Foundry.
Microsoft runs the agent loop and sandbox, while the customer supplies the instructions, skills, tools and model choice.
For businesses, this could make AI development more accessible, particularly for organisations that don't have the resources to build and maintain their own agent infrastructure.
The leading AI model can change quickly, but the integrations, access controls and monitoring surrounding it need to remain reliable. A common harness could reduce the work involved in switching models or using different ones for different jobs.
For smaller businesses, a managed service may be more realistic than building everything from scratch.
However, that doesn't remove the need for governance. Businesses should still understand which model handled each task, what data it received, what the task cost, which tools it could use and how a failed action can be stopped or reversed.
Having access to multiple AI models is useful, provided businesses can retain control over how they're used.
OpenAI and Oracle bring AI closer to everyday business operations
OpenAI and Oracle have expanded their partnership to connect Oracle applications and data with ChatGPT and develop new Oracle products using OpenAI models.
One of the main developments is Oracle AI Workspace, which Oracle is developing to help employees find information and complete approved workflows across areas such as finance, procurement and planning from within ChatGPT.
The announcement also covers Oracle AI Security Mesh for authorised vulnerability detection and remediation, Oracle App Composer for building internal applications connected to approved Oracle data, and an Oracle AI SMB Bundle combining ChatGPT Business, Codex, Oracle applications and recurring Oracle Cloud Infrastructure credits.
Oracle customers in the US can now buy the OpenAI API, Codex and ChatGPT Work through Oracle Marketplace using existing procurement processes and eligible Universal Credits.
Oracle also shared an example from its own recruitment team, which reported reducing market research tasks from two to four days to around 15 to 20 minutes.
That's a significant improvement, although it's one company's reported experience rather than a result every business should expect. The Oracle announcement provides more detail on the products and plans.
For businesses, the opportunity is to use AI within the systems where finance, hiring and operations already happen.
That could save staff time moving between applications and completing repetitive tasks, but it also means businesses need to pay close attention to the information AI tools can access.
Before enabling an integration, organisations should understand which records an AI assistant can read, what changes it can make, where human approval is required and what evidence is retained.
Several of Oracle's announced products are still being developed, so it's worth distinguishing between what's available today and what's planned for the future.
Google's Gemini 4 Carbon pushes further into cybersecurity
Google announced Gemini 4 Carbon on October 28th, just two weeks after Gemini 4 Argon.
Google describes Carbon as its strongest model for long-running tasks and cybersecurity work, with improvements in finding, validating and patching software vulnerabilities.
The company says its internal teams have used the model on engineering tasks that can run for days, continuing to work through problems even when initial approaches fail.
For security teams, that could eventually mean more support with identifying weaknesses and improving the speed of remediation.
However, Carbon isn't currently available as a general business product.
Google is initially providing access to selected cyber defenders through its Fairwind programme while additional testing and safeguards are completed.
When the model becomes more widely available, Google says introductory API pricing will match Gemini 4 Argon at $2 per million input tokens and $10 per million output tokens, with cached input priced at a 95% discount.
The announcement shows how quickly AI capabilities are advancing within cybersecurity and software development.
AI-assisted vulnerability management could become increasingly useful, particularly for organisations managing large or complex IT environments.
But these tools need appropriate controls.
Any testing should take place within authorised environments, with clear permissions and human oversight before changes reach live systems.
The same capabilities that help security teams identify vulnerabilities could create problems if they're used without proper safeguards.
Anthropic's IPO filing reveals the costs behind Claude
Anthropic, the company behind Claude, filed its public registration statement for a proposed stock market listing on October 26th.
While the filing doesn't immediately change how businesses use Claude, it provides a clearer picture of the costs involved in developing and operating large-scale AI services.
According to the SEC filing, Anthropic generated $16.3 billion in revenue during the first half of 2026, compared with $1.2 billion during the same period in 2025.
Its operating loss narrowed to $693 million.
The widely reported $80.7 billion net loss needs some context. Around $72.4 billion relates to changes in the valuation of convertible notes, derivative liabilities and other financing instruments.
Other figures are particularly relevant to businesses using AI services.
Consumption-based fees accounted for $13.1 billion, or approximately 80% of revenue, while subscriptions made up the remaining 20%.
Compute and infrastructure expenses reached $8.945 billion, equivalent to around 55% of revenue.
Anthropic also highlighted its reliance on major cloud providers, including Amazon, Google and Microsoft, alongside the concentration of revenue among a relatively small number of customers and distribution channels.
These figures help explain why AI pricing and usage management are becoming increasingly important.
Even when businesses purchase AI through a familiar monthly subscription, the underlying technology often operates on a consumption-based model.
As AI becomes more integrated into everyday workflows, organisations need to understand how usage affects costs and where they may be dependent on individual providers.
Setting budgets, monitoring consumption and avoiding unnecessary reliance on a single AI supplier could help businesses manage those risks.
What does this week's AI news mean for businesses?
AI is becoming a much bigger part of how businesses operate, and this week's announcements show how quickly the technology is developing.
The opportunity to automate everyday tasks is growing, particularly for businesses where staff spend hours moving between systems, processing information or completing repetitive admin.
But there's no need to rush into every new tool or model that comes along.
Start with a useful process
Identify repetitive tasks where AI could save time, then test whether it genuinely improves accuracy, speed or quality.
Keep control of access
Check which systems and information AI agents can reach, and retain human approval for higher-risk actions.
Understand the costs
Monitor usage and the cost of completed tasks, particularly when working with consumption-based AI services.
Bring employees into the process
Give staff the right training and involve them in identifying tasks where AI can make their working lives easier.
For most businesses, a good starting point is to identify a task that's taking up valuable time and see whether AI can genuinely improve it. Test it properly, understand the costs and make sure the right safeguards are in place before rolling it out more widely.
And don't lose sight of the people using it. The best results will come from giving staff the right tools, involving them in the process and making their working lives easier.
That's where businesses are likely to see the greatest value from AI, regardless of which model is making the headlines next week.
This week's AI news FAQs
What is OpenAI GPT-6.1 Terra?
GPT-6.1 Terra is an OpenAI model designed for visual reasoning and computer-based tasks, including working with forms, documents and existing software interfaces. Its standard API pricing starts at $1 per million input tokens and $5 per million output tokens.
What is the Microsoft Copilot harness?
The Copilot harness is the software layer around an AI model that manages instructions, context, tools and execution. Microsoft has made it available through Foundry, allowing developers to work across multiple AI models while retaining a common agent infrastructure.
What does the OpenAI and Oracle partnership mean for businesses?
The expanded partnership aims to connect ChatGPT with Oracle applications and approved business workflows. It includes planned products for enterprise operations, security, application development and smaller businesses. Availability varies by product and region.
Is Google Gemini 4 Carbon available to businesses?
Not generally. Google is initially providing Gemini 4 Carbon to selected cybersecurity defenders through its Fairwind programme while it completes further testing and safeguards.
Why does Anthropic's IPO filing matter to AI customers?
The filing offers insight into the infrastructure costs, usage-based revenue and cloud-provider dependencies behind large AI services. Businesses can use those insights when reviewing AI budgets, supplier risk and long-term technology choices.
How should UK businesses approach AI automation?
Start with a specific business process, test the technology in a controlled environment, check data access and permissions, and measure whether the outcome justifies the cost. Keep people involved in decisions that affect customers, finances or sensitive information.
Turn AI capability into controlled business value
We can help you identify useful AI workflows, understand the systems and data involved, control access and build governance around what happens next.
Talk to our team
Bring us the workflow, AI tool or business process you're considering and we'll help you work through the practical next step.
Explore AI with Fifosys
Understand how AI can fit into your business with the right foundations, data controls and governance in place.