Gemini Business Gets Claude: Smart, Surprising Model Choice

Gemini Business customers are about to get something Google has never offered inside Gemini before: a model picker that includes Anthropic’s Claude. A launch video for the new Gemini agent, published by Google Cloud chief executive Thomas Kurian on Thursday 8 October 2026, shows a drop-down menu listing Gemini 4 Argon, Gemini 3.8 Flash, Claude Opus 5 and Claude Sonnet 5.5, sitting under an “Auto” option that lets Gemini choose for you.

The detail was first flagged by TestingCatalog on X, which wrote that “it is the first time when Claude models would appear on Gemini, along with Google’s models”. Google’s own announcements confirm the policy behind the picture. Kurian’s keynote post says the agent orchestrates “across our Gemini family of models and Claude models from Anthropic today, and other leading private and open models in the future”.

This article explains what Google showed, how model choice is meant to work in Gemini Business, why a rival’s models inside Gemini matter, and what is genuinely available today versus still in early access. It also covers cost controls, the competition from Microsoft and OpenAI, and the checks a small or midsize firm should run before switching Claude on. For the launch itself, see our earlier report on Google Cloud’s Gemini agent for work.

Table of contents

What Google Showed for Gemini Business
How Model Choice Works in Gemini Business
Why Claude Inside Gemini Business Matters
What Small and Midsize Firms Get From Gemini Business
The Early-Access Fine Print on Gemini Business
Cost Controls Around Gemini Business Model Choice
Gemini Business vs Microsoft Copilot and ChatGPT Work
What Gemini Business Customers Should Check Before Using Claude
What Comes Next for Gemini Business
Frequently Asked Questions About Gemini Business and Claude
References

What Google Showed for Gemini Business

The Gemini agent is Google’s new “single universal agent for work”. It answers questions, does knowledge work, creates content and writes code from one prompt box, and it can run long jobs in the cloud while you are away. Google unveiled it at its Gemini at Work 2026 event, where chief executive Sundar Pichai said Gemini now has more than one billion monthly active users, according to TechCrunch’s report from the event.

The model picker is the part that matters for Gemini Business buyers. We stepped through Kurian’s 1 minute 42 second launch video frame by frame. About 78 seconds in, a title card reads “You choose the models that fit your needs”, and the next shot opens the model menu inside the Gemini prompt box. The closing card points to Google’s “Gemini for business” page.

The four models in the Gemini Business picker

Picker entry	Maker	What it is for	Developer API list price (input / output per 1M tokens)
Auto	Google	Default; the menu says “Gemini chooses the best fit”	Not applicable
Gemini 4 Argon (tagged “New”)	Google	Frontier reasoning, in Kurian’s words	$2 / $10 introductory, then $4 / $20
Gemini 3.8 Flash	Google	Speed and volume	$0.75 / $3.75 to 31 December 2026, then $1.50 / $7.50
Claude Opus 5	Anthropic	The Opus model before Opus 5.5	$5 / $25
Claude Sonnet 5.5	Anthropic	Anthropic’s mid-tier workhorse, launched in late September	$2 / $10

The prices come from the Claude pricing page, the Gemini API pricing page and Google’s Argon launch post, which we covered in our Gemini 4 Argon report. They are developer API rates, not what Gemini Business will charge per seat or per task, and Google has not published that. We include them because they show how wide the cost spread is between the models in one menu.

Who spotted it, and why the naming differs

TestingCatalog’s post went up at 18:55 UTC on 8 October, about 90 minutes after Kurian’s video, and listed the models as “Gemini Argon 4” and “Gemini Flash 3.8”. The menu itself reads “Gemini 4 Argon” and “Gemini 3.8 Flash”, so we use Google’s spelling. TestingCatalog also calls the product “Gemini Business”. Google’s own page now calls the offering “Gemini for business”, and Gemini Business is also the name of the small-company edition of Gemini Enterprise, which Google’s edition comparison lists at one to 500 users with 25 GiB of pooled storage.

One more detail is worth noticing. The menu shows Claude Opus 5, not the newer Claude Opus 5.5 that Anthropic now sells, and Opus 5 is actually the dearer of the two on Anthropic’s API. Launch videos are often recorded weeks before an event, so the shipping Gemini Business picker may list different versions. Treat the four names as a snapshot, not a contract.

How Model Choice Works in Gemini Business

Google’s design separates the agent from the model. Kurian’s keynote post puts it plainly: “Gemini is the agent, and the model underneath it is a separate choice.” The agent holds your memory, your connected tools, your skills and your permissions. The model does the reasoning for one job, and it can change from job to job.

That split is the whole pitch for Gemini Business. If a better model arrives next quarter, Google says your context, skills and data “stay put” while the engine underneath changes. For a small firm that has spent months connecting Gemini to its email, files and accounting tools, not having to rebuild everything for a new model is a real saving.

Auto routing is the default

Out of the box, Gemini Business does not ask staff to pick a model. TechCrunch reported that “by default, the AI will pick the best model to complete the task”. Google calls this smart routing: the agent triages each workload so it “runs on the model that delivers maximum performance at the lowest possible cost”, in the words of Google’s SMB announcement.

The reasoning is sound. Kurian argues that “the best model for the task is not always the largest one”. A quick summary of a meeting does not need a frontier model, while a tricky contract comparison might. Routing the easy work to a cheaper model and the hard work to a stronger one is how Google expects Gemini Business customers to control spend without thinking about it.

Taking over the picker

Staff who want a specific model can override Auto. TechCrunch described it as users being able to “take over to choose a model, including those from third parties, starting with Anthropic’s Claude models”. In the video, a cursor opens the menu and hovers over Claude Sonnet 5.5.

Google has also said the list will grow. TechCrunch reported plans to add “open source models and other private models in the future”, and Kurian’s post mentions “other leading private and open models”. Nothing more specific has been announced, so do not plan a Gemini Business rollout around a model that is not yet on the list.

Ask for approval and other controls in the prompt box

The same shot shows a second control beside the attachment button labelled “Ask for approval”. It appears to set how much the agent may do before checking with you. A later security card in the video lists “Human sign-off on high impact actions”, “Budget caps for cost control” and “Agent gateway for policy enforcement”.

Why Claude Inside Gemini Business Matters

Google builds its own frontier models, runs its own chips and sells its own agent. Putting a competitor’s models into its flagship work product is a notable concession. TestingCatalog’s view was that it “should help Google to compete for enterprise customers as loads of them are moving to Claude and ChatGPT these days”.

There is a simple commercial logic here. If a customer’s staff prefer Claude for writing or coding, Google would rather they use Claude inside Gemini Business than leave for Claude’s own apps. Google keeps the account, the data connections, the admin console and the bill. Anthropic gets its models in front of millions of new business users. Kingy AI made the same point in its analysis of the launch: Google can stay central even when a customer prefers a rival’s intelligence for a particular job.

Not the first time Claude has run on Google Cloud

TestingCatalog is right that this is a first for Gemini itself, but Claude has lived on Google’s infrastructure for a while. Google Cloud has sold Claude models to developers through Vertex AI’s Model Garden since 2024, and Google is an investor in Anthropic. What changes now is the audience. Vertex AI is a developer platform; Gemini Business is an app that office staff, bookkeepers and marketers open every day.

A pattern across the industry

Google is not alone in mixing models. Microsoft added Anthropic’s models to Microsoft 365 Copilot in 2025, and its updated Copilot launched on 25 September 2026. OpenAI, by contrast, keeps ChatGPT on its own models. For buyers, the trend is clear: the agent and the workspace are becoming the product you commit to, and the model is becoming a setting you change.

What Small and Midsize Firms Get From Gemini Business

Google aimed a separate announcement squarely at smaller companies. Sharon Prosser, Google Cloud’s VP for SMB and Scaled, wrote in Google’s SMB post that “for small and midsize businesses, with lean teams and margins, Gemini will be a critical tool”. She said SMB usage of Google Cloud’s AI tools grew more than fivefold year on year.

For a Gemini Business customer, the model picker is one of four headline benefits Prosser lists. The others are connections to common SMB tools, access from every device, and cost controls. Each is worth a closer look, because together they decide whether Claude inside Gemini Business is useful or just a novelty.

Connectors to the tools small teams already use

Prosser names Asana, Box, Clay, Docusign, Dropbox, GitHub, LegalZoom, Notion, Salesforce, Shopify, Slack and Wix. The product page claims “over 100 supported integrations”, including Microsoft 365, Confluence, Jira, Linear, Monday, Smartsheet, GitLab, HubSpot and ServiceNow. Google says data stays where it lives and access follows each user’s existing permissions. If you already run Model Context Protocol servers, see our guide to custom MCP server connections in Gemini Business.

Roles instead of prompts

Google wants Gemini to act like a team member with a defined job. Prosser’s examples are a project manager, a bookkeeper, an inventory manager and a marketing consultant. TechCrunch reported that a coworker agent gets its own Workspace account and email address, so staff can tag it, email it or add it to a group chat.

Model choice for different jobs

This is where Claude inside Gemini Business could earn its place. A firm might let Auto handle routine admin, then pin a marketing role to whichever model writes best for its brand voice, and pin a reconciliation role to whichever model makes fewest arithmetic slips. Because the roles share one memory and one set of connectors, switching a role’s model should not break its access.

The cost spread inside one menu

The chart below shows why routing matters so much to a Gemini Business bill. It compares the developer API list price for one million output tokens across the four models in the picker. It is not Gemini Business pricing, which Google has not published, but the relative gaps are a reasonable guide to what Google pays to run each job.

API list price per 1M output tokens, standard rates (Opus 5 = 100%)

Claude Opus 5 $25

Gemini 4 Argon, after introductory pricing $20

Claude Sonnet 5.5 $10

Gemini 3.8 Flash, 2027 rate $7.50

On those numbers, a job sent to Claude Opus 5 costs about 3.3 times as much per output token as the same job on Gemini 3.8 Flash at its 2027 rate ($25 against $7.50), and about 6.7 times as much at Flash’s current promotional rate of $3.75. That is the gap Auto routing is designed to exploit, and the gap a staff member closes the moment they pin every task to the most expensive model.

The Early-Access Fine Print on Gemini Business

The launch messaging and the small print do not quite agree. Both Kurian’s keynote and Prosser’s SMB post say the agent orchestrates across Gemini and Claude models “today”. But the FAQ on Google’s Gemini for business page answers “Which models can I use?” with “Google Gemini models today, with additional models coming soon.”

The same page carries a disclaimer: “Some capabilities, including delegating multi-step background tasks to Gemini, Gemini on mobile and desktop, and third-party model choice, are currently in early access and will be available to all customers soon.” Prosser’s post also says “Gemini is in early access with SMB customers today and will be available to all customers soon.”

What is live and what is not

Capability	Status Google states	Where it says so
Gemini models in the agent	Available	Product FAQ
Claude models (third-party model choice)	Early access; “coming soon” for everyone	Product FAQ and disclaimer
Multi-step background tasks	Early access	Product page disclaimer
Gemini on mobile and desktop	Early access	Product page disclaimer
The agent for SMB customers	Early access; all customers “soon”	Google SMB blog
Open-source and other private models	Planned, no date	Kurian keynote, TechCrunch
Consumer version of the agent	Later, after businesses	TechCrunch, citing Pichai

Reading the mismatch

The most likely reading is that Claude works today for early-access customers and reaches everyone later. That fits the “early access” wording and the “Claude models from Anthropic today” line in the keynote. But no Google page gives a general availability date for Claude in Gemini Business, and no page we found lists which editions include the picker.

So the honest answer to “can my team use Claude in Gemini Business on Monday?” is: only if Google has put your organisation in early access. Everyone else should ask their Google Cloud account team or reseller for a date in writing.

The free trial does not settle it

Google’s FAQ offers a 30-day trial of the Gemini Enterprise Plus edition “without sales assistance”. That is a good way to test the agent, its connectors and its memory. It does not promise early-access features, so do not assume a trial account will show Claude in the menu.

Cost Controls Around Gemini Business Model Choice

Letting staff choose a model that costs several times more than the default is a budget risk. Google’s answer has three parts, all named in the SMB post: multi-model orchestration, smart routing, and “real-time project-level spend caps in the Cloud Billing Console”.

Spend caps arrived before the agent. In late August Google announced flexible billing and cost controls for agents, including budget alerts at set thresholds. PPC Land reported on launch day that project caps pause Gemini agents when the limit is reached, rather than letting a runaway job keep spending.

Practical guard-rails for a Gemini Business rollout

Leave Auto as the default for everyone, and grant manual model choice only to the roles that need it.
Set a project-level spend cap before the pilot starts, not after the first surprise invoice.
Ask Google how a Claude job is billed: per token at a premium, from a pooled allowance, or inside the seat price.
Review which roles override Auto most often after two weeks, and check whether the results justify it.

Our team helps clients set these limits as part of wider cloud cost optimisation work, and the same principle applies here: decide the ceiling first.

Gemini Business vs Microsoft Copilot and ChatGPT Work

Gemini Business now competes with two other “one agent for work” products that launched within a fortnight. VentureBeat’s launch-day analysis lists the timeline: Microsoft’s updated Copilot on 25 September, OpenAI’s Dots agents on 29 September and Anthropic’s Claude for Google Workspace on 6 October.

How the three compare on model choice

Question	Gemini Business agent	Microsoft Copilot	ChatGPT Work
Models from more than one lab?	Yes: Gemini plus Claude, more planned	Yes: models from OpenAI and Anthropic	No: only models from OpenAI
Default behaviour	Auto routing picks the model	Microsoft picks per feature	User or admin picks a GPT model
Home workspace	Google Workspace, also Microsoft 365 and Slack	Microsoft 365	ChatGPT, with connectors
Agent identity	Own Workspace account and email	Autopilot agent with its own identity	Dots agents work across connected apps
Availability of model choice	Early access	Rolling out by feature	Not applicable

Where Gemini Business has the edge

If your firm lives in Google Workspace, Gemini Business is the shortest path to an agent that already knows your mail, calendar and Drive. Adding Claude removes one of the main reasons staff might have wandered off to a separate tool. The cross-suite reach matters too: TechCrunch listed Microsoft 365, ServiceNow and Slack among the surfaces the agent can run in.

Where it does not yet

Microsoft and OpenAI have more of their agent features generally available today. Google’s biggest differentiator, model choice, is the part still marked early access. Until it ships widely, the comparison above describes Google’s plan more than its product.

What Gemini Business Customers Should Check Before Using Claude

Mixing models from two companies raises questions that a single-vendor tool does not. None of them is a reason to avoid Claude inside Gemini Business, but each should be answered before staff use it with client data.

Data processing and residency

Google’s FAQ promises that prompts and outputs are “never used to train Google models or models for any other customer”. Ask whether that commitment, and your existing data processing terms, cover requests that Gemini routes to Claude. Ask where those requests are processed, and whether Anthropic acts as a sub-processor or Google runs the model inside its own cloud as it does on Vertex AI. UK firms should record the answer in their data protection impact assessment.

Admin controls and audit trails

Check whether administrators can restrict which models each role or user may select, and whether audit logs record which model produced each output. Agent-attributed audit trails are part of Google’s pitch, so the model name should be in them. This is basic cybersecurity hygiene for any agent that can act on your systems.

Regression testing when models change

A workflow tuned on one model can behave differently on another. Kingy AI warned that a model change “should therefore trigger regression checks even when the agent interface looks unchanged”. Keep a small set of real tasks, such as a standard invoice query or a weekly report, and rerun them whenever a role’s model changes.

A short pilot checklist

Confirm in writing whether your organisation has early access to Claude in Gemini Business.
Pick two roles and run the same ten tasks on Auto, on a Gemini model and on a Claude model.
Record time saved, corrections needed and cost per accepted result, not just speed.
Set a spend cap and a human sign-off rule for anything that sends email or changes records.
Decide who may override Auto, and review that list monthly.

If you need help shaping an AI strategy around a multi-model agent, start with the tasks, not the models.

What Comes Next for Gemini Business

Three things are worth watching. First, a general availability date for third-party model choice, which would turn this from a demo into a feature every Gemini Business customer can use. Second, the first open-weight models in the picker, which Google has promised without naming. Third, pricing: whether Google charges more when a user picks Claude, or absorbs the difference inside the seat.

Google’s own models will keep moving too. Gemini 4 Argon carried a “New” tag in the menu, and Kurian’s keynote also named Omni for generative media and Gemma for lightweight edge work as part of Google’s model stack. For our running coverage of model releases, visit the AI models and tools hub.

The bigger story is that the model is becoming a setting rather than a commitment. For a small business, that is good news. It means the work you put into connecting an agent to your systems is less likely to be wasted when the next model wins the benchmarks.

Frequently Asked Questions About Gemini Business and Claude

Can I use Claude in Gemini Business today?

Only if Google has given your organisation early access. Google’s announcements say the agent works with Claude models “today”, but its product FAQ lists Gemini models today and “additional models coming soon”, and calls third-party model choice early access.

Which Claude models will Gemini Business offer?

The launch video shows Claude Opus 5 and Claude Sonnet 5.5, alongside Gemini 4 Argon and Gemini 3.8 Flash. Google may update the list before general release, since Anthropic has since launched Claude Opus 5.5.

Do I have to choose a model in Gemini Business?

No. The default is Auto, where Gemini chooses the best fit for each job. Picking a model manually is optional.

Does using Claude cost more in Gemini Business?

Google has not said. On the developer APIs, Claude Opus 5 costs more per token than Gemini 3.8 Flash, so ask your account team how manual model choice is billed before you enable it widely.