What Changed in AI This Week? ChatGPT Business Profiles, Claude Fable 5.2, Copilot Scale and Gemini 4 Carbon
OpenAI turned ChatGPT into a business discovery channel, Anthropic and Google launched specialist models for harder work, and Microsoft put the focus back on the awkward bit: moving from pilots to measurable value.
It's been another busy week in AI, with announcements from OpenAI, Microsoft, Anthropic and Google that could have a real impact on how businesses use the technology.
OpenAI is making it easier for customers to find businesses through ChatGPT, while its latest partnership with AWS could simplify how larger organisations purchase AI tools. Anthropic and Google have released more powerful models designed for complex work, and Adobe is bringing more of its tools into Google's ecosystem.
Microsoft has also shared some revealing figures about AI adoption. Despite growing investment, many businesses are still struggling to turn early experiments into something that delivers consistent value.
Now, with more AI tools becoming available and more ways to connect them to everyday systems, businesses have plenty of options. But choosing where to invest, understanding what actually works, and getting staff on board remain among the biggest challenges.
Headlines at a glance
OpenAI launches Businesses in ChatGPT
Eligible companies can claim a profile, update information and see aggregated discovery insights. Business Agents begin with eligible US ecommerce and online service firms.
ChatGPT Enterprise arrives in AWS Marketplace
AWS customers can buy and manage it through existing purchasing and billing relationships and may apply eligible spend to AWS commitments.
Anthropic releases Claude Fable 5.2 and Mythos 5.2
The specialist models target demanding technical work, with faster output and lower API pricing than their predecessors.
Microsoft puts the scale gap in plain view
Its latest update says only 48% of organisations in sponsored IDC research are scaling AI or seeing consistent value, despite spending plans continuing to rise.
Google launches Gemini 4 Carbon and expands its Adobe partnership
A new specialist model targets long-running technical work while Adobe tools move closer to Gemini, Gmail and Google Drive.
Zooming out a tad, there's a reasonably clear thread running through those announcements. The market is splitting between broad, everyday AI and more expensive specialist systems, while distribution is moving into the places where people already search, buy and work.
ChatGPT is becoming a business discovery channel
OpenAI has introduced Businesses in ChatGPT in beta. Through a new Business Hub, eligible organisations can claim and verify their business, update core details, and add product feeds that include pricing and availability.
The more interesting bit, though, is the visibility data. OpenAI says businesses can see aggregated figures for mentions and website clicks, along with "discovery topics": common themes in conversations where the company's website was cited. It says individual conversations are not shared and that insights only appear once activity meets minimum thresholds.
That starts to make ChatGPT look less like a separate AI tool and more like another customer acquisition surface. Search visibility has never been entirely under a company's control, of course, but businesses now have another profile to claim, another information feed to keep accurate and another source of referral traffic to measure. Welcome to the admin list.
There is an important availability caveat, as OpenAI says eligible businesses in locations where ChatGPT is available can use Business Hub, but Business Agents are initially limited to eligible ecommerce and online service businesses in the United States. Those agents can answer follow-up questions about a company's offering and can also be used as a destination for clearly labelled ChatGPT advertising. UK businesses should therefore separate what they can do now from what remains a US-first product rollout.
Claim the profile if eligible, then treat it like any other public listing: verify the organisation, use an accountable business email, check product and service information, and assign an owner for updates. It is also worth reviewing whether your website gives AI systems clear, current and consistent facts about what you sell, where you operate and how customers can contact you.
Don't rush to optimise every page for a mysterious new acronym. Start with sound content hygiene, then compare ChatGPT mentions and clicks against existing analytics. If the numbers are too small to appear, don't worry. That's still useful information.
ChatGPT Enterprise is now available through AWS Marketplace
A day earlier, OpenAI made ChatGPT Enterprise available in AWS Marketplace. Organisations can purchase and manage it through an existing AWS billing relationship, with eligible spending counting towards AWS commitments.
This is more procurement news, rather than a new AI capability, but that doesn't make it any less trivial. Larger growing businesses often lose weeks to vendor onboarding, security review, contract routing and separate billing arrangements. Marketplace purchasing can reduce some of that friction and consolidate spend, particularly where an AWS agreement is already in place.
It also makes the platform distinction clearer, with OpenAI models in Amazon Bedrock used by developers building applications and agents, and ChatGPT Enterprise as the managed workspace used directly by staff. A business may use one, both or neither. They solve different problems and shouldn't be treated as interchangeable simply because both now sit inside the AWS commercial ecosystem.
If your organisation already has an AWS commitment, compare marketplace procurement with a direct OpenAI agreement in terms of total cost, contract terms, data handling, support, and exit options. Consolidated billing is convenient, but it is not a substitute for checking which product your teams actually need or how access will be governed.
Anthropic launches Claude Fable 5.2 and Mythos 5.2
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, two versions of the same underlying model aimed at demanding technical work across coding, research, cybersecurity and science.
Fable is the broadly available version, with safeguards for cyber and biological misuse. Mythos is reserved for approved organisations through Anthropic's verification programmes. Anthropic says the models generate output up to 30% faster than Fable and Mythos 5.1 and cost $8 per million input tokens and $40 per million output tokens. An optional fast mode doubles those token prices in return for output up to 2.5 times faster.
The part of the announcement you should pay attention to is Anthropic's own product guidance, which recommends Opus 5.5 as the everyday starting point and positions Fable or Mythos for the hardest, more open-ended work. In other words, the newest model is not automatically the sensible default. Fable can also act as an adviser to smaller models or orchestrate them, which reflects a wider shift towards using several models within one workflow rather than paying the premium rate for every step.
Anthropic published an extensive set of benchmark and customer results, but businesses should treat vendor benchmarks as evidence to investigate, not a purchase order. The relevant test in the coming days and weeks ahead will be on whether the model improves accuracy, completion rate or review time on your own high-value tasks.
Don't swap the default model across the organisation because a benchmark chart moved. Identify the few workflows where difficult reasoning or long-running technical work creates real value, test those against a known baseline and track the full cost per completed task. Faster output is useful; paying twice the token rate for fast mode only makes sense when elapsed time has a genuine business cost.
Microsoft says the real AI problem is scaling value
Microsoft didn't announce a major new Copilot product this week... For once! Its more useful contribution was a business update on moving from AI ambition to operating impact. Microsoft says its Frontier Company teams have worked on nearly 500 projects with more than 200 customers in 28 countries.
The figures Microsoft highlighted from IDC research are more revealing, with 66% of surveyed organisations reporting substantial or transformative AI results; only 48% said they were scaling AI or seeing consistent value across the organisation. At the same time, 71% expected AI spending to rise by at least 10% over the next year. The research was sponsored by Microsoft, so it should probably be read through that lens, but regardless of any potentially skewed findings, the mismatch between enthusiasm and budget, with operating discipline often moving faster, is somewhat familiar.
Microsoft's argument is that value closes workflow by workflow, with business experts working alongside engineers, which, conveniently, is also the bit many smaller organisations can apply without needing to create a 6,000-person transformation unit. Pick a process, give it an owner, define the acceptable inputs and outputs, and measure what changed. Buying Copilot seats for everyone and hoping useful habits emerge is not an operating model.
Review AI projects using four numbers: adoption, task completion, time saved after checking, and total cost. Add a qualitative check for risk and user experience. If a pilot cannot show which workflow changed and who owns it, expanding the licence count will probably expand the ambiguity as well.
Google launches Gemini 4 Carbon as Adobe moves closer to everyday work
Google released Gemini 4 Carbon for paying developers and enterprises through Google AI Studio and Gemini for business. Google positions Carbon as a specialist model for long-running, complex work in software engineering, cybersecurity, research and science, while recommending Gemini 4 Argon for most everyday knowledge and coding tasks.
The introductory API price is $2 per million input tokens and $10 per million output tokens, which is no different from Gemini 4 Argon. Google also previewed a swarm feature in its Antigravity development environment, where Carbon can coordinate faster or lighter models across parallel work. Again, the direction is towards model portfolios and orchestration rather than one expensive model doing absolutely everything.
The more immediately visible workflow news came from Google and Adobe's expanded partnership. Adobe Acrobat links are due to appear in Google Drive and Gmail; Adobe is expanding connectors in Gemini and Gemini Enterprise; and Adobe's CX Enterprise Coworker is planned for Gemini Enterprise. Adobe also said its creative agent is expanding across Firefly and Creative Cloud, with connectors and partnerships spanning several major AI platforms.
For marketing and operations teams, that integration story may mean fewer exports and app switches, which can make work faster, but it also creates new routes between documents, prompts, brand assets, and customer data. Convenience increases the importance of checking permissions, retention, and which product processes the content at each step.
If your teams use Google Workspace and Adobe, map the new connectors before enabling them widely. Start with a low-risk workflow such as drafting campaign variants from approved assets. Confirm which data crosses service boundaries, who can publish the result and how brand, accessibility and factual checks are completed.
Quick answer: what should SMEs do about AI this week?
Claim your business profile
Claim and verify your ChatGPT business profile if eligible, then assign somebody to keep it accurate.
Review AI procurement
Separate AI procurement from AI use cases. AWS Marketplace may simplify buying ChatGPT Enterprise, but it does not decide which teams need it.
Test specialist models
Route specialist models to specialist work. Test Fable 5.2 or Gemini 4 Carbon on difficult, valuable tasks rather than making them the default everywhere.
Measure one workflow
Turn one Copilot or agent pilot into a measured workflow with an owner, baseline, review step and budget.
Check new connectors
Review new Adobe, Google and other app connectors as data integrations, not harmless shortcuts.
Fifosys view
There's plenty for businesses to get their heads around this week, particularly as AI becomes more closely connected to the tools their staff and customers already use.
Being discoverable through ChatGPT could create new opportunities to reach customers. Better integrations could save staff time, and more capable models could help with work that's previously been difficult to automate.
But businesses don't need to adopt everything as soon as it's released.
For many, there's probably more value in making better use of the tools they've already invested in. That might mean revisiting a Copilot rollout, finding a repetitive task that could be automated or simply making sure staff understand what AI tools they're allowed to use.
It's also worth involving the people doing the work. They'll often have the clearest idea of where time is being wasted, which processes cause frustration and where AI could genuinely help.
The technology will keep changing, and there'll be another batch of announcements next week. Businesses that take the time to understand what works for their people and their processes will be in a much better position to benefit from it.
This week's AI news FAQs
What is Businesses in ChatGPT?
It is an OpenAI beta that lets eligible organisations claim a business profile, update information and view aggregated discovery insights in Business Hub. Business Agents initially serve eligible US ecommerce and online service businesses.
Can UK businesses use ChatGPT Business Agents?
Not at launch. OpenAI says Business Agents initially cover eligible businesses in the United States, although eligible organisations elsewhere may be able to claim a profile through Business Hub where ChatGPT is available.
What changed with ChatGPT Enterprise and AWS?
ChatGPT Enterprise can now be purchased through AWS Marketplace using an organisation's existing AWS billing relationship. Eligible spend may count towards AWS commitments.
How much do Claude Fable 5.2 and Gemini 4 Carbon cost?
Anthropic lists Claude Fable 5.2 at $8 per million input tokens and $40 per million output tokens, with fast mode priced at $16 and $80 respectively. Google lists Gemini 4 Carbon at an introductory $2 per million input tokens and $10 per million output tokens. Usage patterns and task completion rates matter more than token price alone.
What should businesses measure in an AI pilot?
Measure real use, successful task completion, time saved after review, total licence and usage cost, and any change in risk or customer experience. Tie the pilot to a named workflow and owner before scaling it.
Make AI useful, secure and measurable
From choosing the right tools to connecting them safely with your existing systems, we can help you take a practical approach to AI adoption.