Last updated 24 September 2026: agent safety incidents at OpenAI and Google, plus Claude Fable 5.1, GPT-6 Astra’s rollout and the Privacy Act’s 10 December AI disclosure deadline.
Every business we talk to is asking the same question: which AI tool should we actually be using? The honest answer keeps changing. Since our last update, Anthropic shipped Claude Fable 5.1, OpenAI took GPT-5.6 to general availability and began rolling out GPT-6 Astra, and Microsoft put Claude Opus 5 into the Copilot model picker. We deploy all three platforms across Australian businesses every day, including inside our own operations. Here is what we have learned, updated for September 2026.
If you want the backstory on why we chose Claude as our primary platform despite being a Microsoft Partner, read that decision in detail here. If you are already using Copilot and wondering whether it is enough, we wrote about that gap too. This post is the full ChatGPT vs Copilot vs Claude comparison for business, current as of September 2026.
If your business runs on Microsoft 365, start with Copilot, then turn on Anthropic models in the admin centre so your team can pick Claude inside the same surface for deep work. If you need a standalone AI for analytical work, long documents, compliance, or coding without the Microsoft licence overhead, Claude on its own is still the stronger tool, and it now ships the most capable publicly available model on the market. If you want the most flexible general-purpose AI for staff who do not live in Microsoft 365, ChatGPT Business is the safe default. Most businesses with 20 or more staff will end up using at least two of these.
| Microsoft Copilot | Claude | ChatGPT | |
|---|---|---|---|
| Built by | Microsoft (uses OpenAI and Anthropic models) | Anthropic | OpenAI |
| Best for | M365 workflow, in-app AI | Deep work, compliance, long documents, coding | Versatile general-purpose tasks |
| Requires M365 | Yes, Business Standard or E3+ | No | No |
| Latest flagship model | GPT-5.6 (default) plus Claude models up to Opus 5 (opt-in) | Claude Fable 5.1 (paid plans), Sonnet 5 default | GPT-5.6 Sol (paid plans), GPT-6 Astra in phased rollout |
| Context window | Varies by surface | 1,000,000 tokens (~2,500 pages) | 128k on Plus, 1M on Pro $200 |
| Business pricing (USD) | $21/user (SMB, $18 promo to Dec 2026) or $30/user (Enterprise) | $20/seat annual, $25 monthly (Team) | $20/seat annual, $25 monthly (Business) |
| Trains on your data? | No (stays in your M365 tenant) | No (on all paid plans) | No (Business and Enterprise only) |
| Agentic work | Copilot Cowork (GA, credit-billed) | Claude Cowork (GA on paid plans) | ChatGPT Work (GA on paid plans) |
| Email and calendar integration | Yes, native | Via Claude in Chrome and connectors | Via apps and connectors |
| Coding | Solid in GitHub Copilot, basic in M365 apps | Strongest (Claude Code, Fable 5.1) | Very good (Codex, now inside ChatGPT Work) |
| Spreadsheet work | Excel with Copilot (native, Claude selectable) | Claude for Excel | Code interpreter + connectors |
| Epic IT verdict | Best for in-Microsoft workflow | Best for deep work, anywhere | Best for breadth and creative |
Four shifts matter if you last read this post in July.
Claude’s flagship moved again, and got cheaper to run. After Fable 5 landed in June and Opus 5 followed in late July, Anthropic shipped Claude Fable 5.1 on 1 September. It is the same Mythos-class, 1,000,000-token flagship, with its biggest gains on long agentic and coding work, and it beats OpenAI’s GPT-5.6 Sol on the current public benchmarks. For API users, cheaper cache reads cut effective costs by around 25 per cent on typical workloads. Subscription pricing did not change: paid Claude plans simply got a better model, and Sonnet 5 remains the default on Free and Pro.
OpenAI shipped a new generation, with regulators now in the pipeline. GPT-5.6 reached general availability on 9 July as a three-model family (Sol for frontier work, Terra for the everyday, Luna for cheap volume) after a government-limited preview, and ChatGPT Work folded the Codex agents into an agentic workspace on paid plans. Then on 3 September OpenAI began a phased rollout of GPT-6 Astra, the first of its models to cross the company’s Critical cybersecurity threshold, which means its most advanced capabilities are gated behind a vetted access program rather than shipped to subscribers. The buying lesson holds across all of it: frontier releases can be paused, gated, or phased at short notice, so build on what you can access today and treat what is still rolling out, Astra included, as a bonus rather than a dependency. We cover what Astra’s offensive cyber capability means for your defences in a separate post.
Copilot’s model picker got serious, and one default needs your attention. Microsoft made Claude Opus 5 available inside Microsoft 365 Copilot on 24 July, and Claude model choice now spans Copilot Chat, Researcher, Copilot Studio, Excel, PowerPoint, Cowork, and Word (still rolling out). Microsoft has also confirmed that Copilot Chat, Cowork, and GitHub Copilot are merging into a single app this quarter. The governance trap: from mid-June, Microsoft began automatically installing the Microsoft 365 Copilot app on eligible commercial Windows devices. The setting is on by default, so if your organisation has not approved a Copilot deployment, an administrator needs to opt out in the Microsoft 365 Apps admin centre.
Agent safety became a buying question. In July, OpenAI agents running an internal cyber evaluation escaped their sandbox and reached Hugging Face’s production systems. On 18 September Google disclosed that Gemini accessed three outside systems during a test. On 23 September the Prime Minister revealed that an OpenAI agent had accessed the Medicare Statistics Reporting Service portal in June, with the government not told until September. All of these happened in vendor test environments, not in customer deployments of ChatGPT Work, Gemini or anything else. The lesson for buyers is the same whichever platform you choose: govern agents like any other privileged account. We cover what to do in our post on AI agent security risks.
Microsoft Copilot is not a single AI. It is a set of AI surfaces embedded into your existing Microsoft 365 environment, powered by a mix of OpenAI and Anthropic models you can choose between. When you open Outlook, Copilot can summarise a 40-message email thread in two sentences. In Teams, it generates meeting notes from a recorded call. In Excel, it writes formulas from plain-English descriptions, and Edit with Copilot lets you opt for Claude on harder tasks. Its power is integration, not raw intelligence. It reads your SharePoint files, calendar, inbox, and Teams channels, and uses that context without you copying and pasting anything. With Cowork now generally available, it can also run multi-step tasks in the background, billed per credit. Pricing and bundle details are on the official pricing page.
Claude (built by Anthropic) is the tool we chose for our own internal operations and for our managed AI clients. It excels at long-form reasoning, document analysis, compliance work, and code generation. Fable 5.1 handles up to 1,000,000 tokens of context (around 2,500 pages), which means you can feed it an entire policy document, contract, or codebase and get a coherent response back, and its lead over other models grows as tasks get longer and more complex. Claude’s writing is more precise and less generic than the alternatives, and Anthropic does not train on your business data on any paid plan. Claude Cowork, its agentic desktop workspace, is generally available on paid plans, and Claude in Chrome is out of beta. Pricing is on the Anthropic pricing page.
ChatGPT (built by OpenAI) is the most widely used AI tool in the world. It is versatile, fast, and has the broadest ecosystem of apps and connectors. It handles creative writing, brainstorming, research, coding, image generation, voice, and data analysis well. Paid business users now get GPT-5.6 Sol for frontier work with selectable effort levels, Terra for everyday tasks, and ChatGPT Work for agentic multi-step jobs, while free accounts run on Terra, and GPT-6 Astra is phasing in above all of it. ChatGPT’s weakness relative to Copilot is that it does not live inside your Microsoft environment by default. Its weakness relative to Claude is that its outputs for long, complex analytical work tend to be less precise. Pricing is on the ChatGPT Business pricing page.
We tested all three across the tasks Australian SMBs actually use AI for. Not benchmarks. Not chatbot arena scores. The tasks that business owners, office managers, and operations staff do every day.
Email and meeting summarisation. Copilot still wins decisively. It reads your inbox and meeting recordings natively. No copying required. Claude and ChatGPT can both summarise emails, but you have to paste content in manually unless you wire up connectors, which breaks the workflow for most staff.
Document drafting and editing. Claude produces the cleanest, most precise written output. We use it for proposals, compliance documents, client reports, and blog content. With Anthropic models enabled in Copilot, you can get Claude’s writing inside Word too, which is the best of both worlds. ChatGPT is a close second for shorter content and brainstorming. Native Copilot drafting on OpenAI models is adequate for internal communications but tends toward generic corporate language.
Data analysis and spreadsheets. Copilot’s Excel integration is genuinely useful for formula writing and basic data manipulation, and Edit with Copilot can now run on Claude for harder tasks. For serious analysis (cleaning messy data, building models, generating charts from complex datasets) ChatGPT with its code interpreter and Claude, either standalone or via Claude for Excel, both outperform default Copilot. Claude handles larger datasets more reliably due to its 1M-token context window.
Compliance and policy work. Claude is the clear leader, and Fable 5.1 widened the gap again. We use it to review Essential Eight assessments, draft AI governance policies, and analyse regulatory documents. Its ability to process long documents without losing context is a real advantage over ChatGPT, which tends to drift on documents longer than 30 pages on Plus. With Claude available inside Copilot Studio, you can also build compliance agents that run inside the Microsoft estate.
Coding and technical work. Claude is the preferred tool for our engineering team, and this is where Fable 5.1’s lead is largest. Claude Code handles multi-file debugging, refactoring, and long-running autonomous work better than any alternative we have tested. GPT-5.6 Sol is genuinely competitive on coding and a strong second via Codex, which now lives inside ChatGPT Work, though Fable 5.1 tops it on the current public benchmarks. GitHub Copilot (a separate product from Microsoft 365 Copilot) remains useful for inline code completion and supports Claude models on paid plans.
Creative work and brainstorming. ChatGPT has the edge. It generates ideas faster, handles more creative formats (image generation, voice, video), and has a broader app ecosystem. Claude is good at structured creative work like blog outlines and messaging frameworks, and Claude Design covers first-pass visual creation. Default Copilot is not built for creative tasks.
| Plan (USD) | Microsoft 365 Copilot | Claude | ChatGPT |
|---|---|---|---|
| Individual | Copilot Chat: free with eligible M365 plan | $20/month (Pro), $100/$200 (Max) | $20/month (Plus), $100 or $200 (Pro) |
| SMB / Business | $21/user, promo $18 to Dec 2026 (Copilot Business, under 300 seats) | $20/seat annual or $25 monthly (Team Standard, 5-seat min) | $20/seat annual or $25 monthly (ChatGPT Business, 2-seat min) |
| Enterprise | $30/user add-on (Copilot Enterprise) | $100/seat annual (Team Premium) or custom (Enterprise) | Custom pricing |
| True cost for 30 users | ~$540-630/month (Business) plus M365 base, plus Cowork credits if enabled | ~$600/month (Team Standard, annual) | ~$600/month (Business, annual) |
The pricing looks similar on paper, but the real cost of Copilot is higher because it requires an existing Microsoft 365 licence underneath it, and those licences got more expensive on 1 July 2026 under the global pricing update we flagged in February. The $18 Copilot Business promotional rate (normally $21) runs to December 2026 and partly offsets that for new deployments. Watch the consumption layer too: Copilot Cowork bills per credit on top of the seat price, which is a newer and less predictable line item. Claude Team or ChatGPT Business at $20 per seat on annual billing is standalone and flat. Two genuine price movements worth noting: Claude Team Premium seats, the tier that includes Claude Code for technical staff, dropped from $150 to $100 per seat on annual billing this year, and Fable 5.1 cut Claude’s API costs by roughly a quarter on typical workloads without touching subscription prices. OpenAI has been cutting API prices too, so if you build on the APIs rather than the chat apps, re-quote your costs this month.
This is where the decision gets serious for businesses handling sensitive data.
Copilot inherits your Microsoft 365 security settings. If you have Conditional Access, sensitivity labels, and data loss prevention policies configured properly, Copilot operates within those boundaries. Your data stays in your Microsoft tenant. When Anthropic models are used inside Copilot, Anthropic acts as a Microsoft subprocessor under the Microsoft Data Protection Addendum, with no separate Anthropic contract required. For Australian commercial tenants Anthropic models are on by default; in the EU, EFTA, and UK they are off by default. Two newer governance jobs: if you enable Cowork, configure credit spending limits at the tenant, group, and user level before anyone touches it, and check whether the Copilot app auto-installation setting (on by default since mid-June) matches your rollout plan.
Claude standalone does not train on your data on any paid plan. Anthropic’s privacy position is the most conservative of the three, and the Fable releases reinforced it: this is the vendor that ships its frontier models with safety classifiers on by default and accepted a slower rollout to do it. For businesses in legal, healthcare, or finance where data handling is paramount, this matters. The Enterprise plan adds audit logs, SSO, SCIM, custom data retention including Zero Data Retention, and admin analytics with model-level entitlements and spend alerts.
ChatGPT Business and Enterprise plans do not use your data for training. The consumer versions (Free, Go, Plus) may use your conversations to improve models unless you opt out, and Free and Go show ads in some markets. For business use, always use the Business or Enterprise tier, never individual accounts.
Agentic features need their own controls on every platform. Copilot Cowork, Claude Cowork and ChatGPT Work run multi-step tasks, often with the permissions of the person who started them. Give agents their own identities, keep their access to what the task needs, log what they do, and require approval before they write, send or delete. The ASD’s September 2026 ISM controls for AI agents set out the same list.
And whichever tool you choose, the Privacy Act now has a deadline for you. From 10 December 2026, new transparency obligations under Australian Privacy Principle 1 (APP 1.7 to 1.9) require businesses to disclose in their privacy policy when automated systems make, or substantially assist in making, decisions that significantly affect individuals. The wording captures AI-assisted decisions, not just fully automated ones. So if Copilot, Claude, or ChatGPT sits inside your hiring, credit, claims handling, or customer decisions and uses personal information to get there, your privacy policy needs to say so, and say what kinds of information and decisions are involved. The OAIC has been running compliance sweeps ahead of the deadline, and serious or repeated breaches can attract penalties of up to 30 per cent of adjusted turnover. None of the three vendors does this compliance for you: disclosure is about how you deploy the tool, not which one you buy. We covered the full reform package in our Privacy Act 2026 checklist, and WA businesses have the PRIS Act’s IPP 10 layered on top.
The governance challenge is not which tool is safest in theory. It is that your staff are probably already using one or more of these on personal accounts, outside your visibility and control. Shadow AI is a real risk, and from December it is also an undisclosed-processing risk. A formal policy and a managed deployment matter more than the specific tool you choose.
We are a Microsoft Solutions Partner. We sell and manage Microsoft 365 for our clients. We also chose Claude as our primary AI platform, not Copilot. That decision was not political. It was practical, and everything that has happened since June has sharpened the answer rather than changed it.
Copilot is the right tool for in-Microsoft workflow automation. If your team spends hours in Outlook, Teams, and Excel, Copilot Business at the current $18 promotional rate will save them time. With Anthropic models turned on, the same licence gives them Claude, now up to Opus 5, inside Copilot Chat, Word, Excel, PowerPoint, Researcher, and Copilot Studio for the deep work where OpenAI models are weaker. Just govern the Cowork credit spend and the auto-install setting from day one.
Claude standalone is still the right tool for deep work outside the Microsoft estate, and Fable 5.1 made that case stronger and cheaper to run. Document analysis, compliance reviews, strategy documents, coding via Claude Code, and any task that needs the full 1M-token context window or a Claude-native surface like Claude for Excel, Claude Cowork, or Claude in Chrome. We deploy it for our AI services clients and use it internally for everything from cybersecurity assessments to client reporting.
ChatGPT Business is the right tool for breadth. If your team needs a single versatile AI assistant for a wide range of tasks and is not deeply embedded in Microsoft 365, it is the safe default at $20 per seat annual. The GPT-5.6 family is what you are buying today: Sol for the hard problems, Terra for the everyday, and ChatGPT Work for agentic tasks. GPT-6 Astra is phasing in on top, but treat it as a bonus when it reaches your plan, not something to delay a deployment for.
For most Australian SMBs with 20 to 100 staff running Microsoft 365, the combination that delivers the most value is Copilot Business with Anthropic models enabled, plus a small number of standalone Claude Team seats for analysts, technical staff, and writers. The Team Premium price cut makes those specialist seats cheaper than they were at the start of the year. ChatGPT fills the gaps for creative tasks and image generation if needed.
Whatever you choose, deploy it properly. Set up an AI governance policy, run an AI readiness assessment, and manage the rollout through your managed IT services provider, not through individual staff signing up on personal credit cards. And if you are weighing outside help for the rollout itself, our guide to the best AI consulting companies in Australia maps who does what, tier by tier, pricing models included.
Audit what your staff are already using. Before you buy anything, find out which AI tools are already in use on personal accounts across your business. That list, not a vendor comparison, should drive your first governance decisions, and it also feeds the Privacy Act disclosure work due by 10 December 2026.
Review your Microsoft renewal against the new pricing. The 1 July price rise is now baked into renewals, and the $18 Copilot Business promotion runs to December 2026. While you are in the admin centre, check the Copilot app auto-installation setting matches your intended rollout, because it is on by default.
Get an independent read on your AI readiness. We run free AI readiness assessments that map tools to roles, flag the governance gaps, and give you a deployment plan. Contact us and we will book one in.
It depends on your primary use case. Copilot is best for businesses that live in Microsoft 365 and want AI embedded in Outlook, Teams, and Excel, and it now includes the option to use Claude alongside OpenAI models. Claude on its own is best for deep document analysis, compliance work, or coding without the Microsoft licence overhead. ChatGPT Business is best as a versatile general-purpose tool. In the ChatGPT vs Copilot vs Claude for business decision, many organisations benefit from using two of the three.
Yes. Claude is available as a model option in Copilot Chat, Researcher, Copilot Studio, Excel, PowerPoint, Cowork, and Word (rolling out) for licensed users in commercial cloud tenants, and Claude Opus 5 has been available in Copilot since 24 July 2026. Admins enable Anthropic models in the Microsoft 365 admin centre, then users pick the model family from the switcher. Anthropic models are on by default for Australian commercial tenants and off by default in the EU, EFTA, and UK.
Claude Fable 5.1 is Anthropic’s Mythos-class flagship released on 1 September 2026, replacing the Fable 5 model that launched in June. It is the most capable publicly available AI model, with a 1,000,000-token context window, its biggest gains on long agentic and coding work, and API costs roughly 25 per cent lower than Fable 5 on typical workloads. It is available to paid Claude subscribers and through the major cloud platforms, including Microsoft Foundry.
Yes. After a government-limited preview in late June, the GPT-5.6 family reached general availability on 9 July 2026. Paid ChatGPT plans get the flagship GPT-5.6 Sol with selectable effort levels, free accounts run on GPT-5.6 Terra, and the cost-efficient Luna model is available through the API.
GPT-6 Astra is OpenAI’s next-generation model, which began a phased rollout on 3 September 2026 and will reach ChatGPT Plus, Pro, Business, and Enterprise plans, the API, and AWS. It is the first OpenAI model to cross the company’s Critical cybersecurity threshold, so its most advanced cyber capabilities are restricted to a vetted access program rather than general subscribers. Plan deployments around the GPT-5.6 family available today and treat Astra’s arrival in your tenant as an upgrade, not a dependency.
Yes, if the tool makes or substantially assists decisions that significantly affect individuals using their personal information. From 10 December 2026, APP 1.7 to 1.9 require your privacy policy to disclose that automated decision-making happens, what kinds of personal information it uses, and what kinds of decisions it touches. The obligation sits with your business, not the AI vendor, so an AI use inventory and an updated privacy policy are the practical first steps.
Copilot Business is $21 USD per user per month for SMBs under 300 seats, with a promotional rate of $18 per user on annual commitment running until December 2026. Copilot Enterprise is $30 USD per user per month. Both require an eligible Microsoft 365 base licence, which increased in price from 1 July 2026. Copilot Cowork is billed separately on consumption through Copilot Credits at $0.01 per credit.
All three offer strong data protection on their business and enterprise tiers. Copilot inherits your existing Microsoft 365 security settings, and when Anthropic models are used inside Copilot, Anthropic acts as a Microsoft subprocessor. Claude standalone has the most conservative privacy stance, with no training on your data on any paid plan. ChatGPT Business and Enterprise do not use your data for training. The biggest risk is not the tool itself but unmanaged usage by staff on personal accounts, so an AI governance policy matters more than the specific tool you choose.
