Claude vs ChatGPT for UK SMEs: a 2026 buyer's guide from Team to Enterprise

Two things have changed since this guide was first written, and both matter more than the model names. Anthropic and OpenAI have converged: the seat prices are within a few pounds of each other, both flagship models now carry a context window of roughly a million tokens, and both vendors hold the same core compliance certifications. What separates them for a UK buyer is no longer capability. It is what each plan tier includes, what currency you are billed in, and a residency position that most procurement documents get wrong.

Prices, plan contents, model line-ups, residency and certifications were re-verified against both vendors' own pages on 10 October 2026, after each released a new model generation, and each claim is labelled with its source. The Cyber Essentials and public-sector points were verified on 17 September 2026. Where a vendor's own surfaces disagree with each other, both are quoted rather than reconciled.

Is Claude better than ChatGPT for a UK business?

Neither is better across the board. For a 10 to 50 seat firm the seat prices are within a few pounds of each other: Claude Team standard seats are £18 a seat a month on annual billing and ChatGPT Business standard seats are £15 (Anthropic's UK pricing page and OpenAI's UK plan picker, 10 October 2026), although the two do not state the same VAT basis. Neither vendor offers UK in-region processing.

The difference is fit. Claude is the stronger choice for coding and agentic workflows, for very long documents (Anthropic bills the full 1M-token window at standard rates, while OpenAI charges twice the input rate and 1.5 times the output rate on GPT-6.1 Sol requests above 272K input tokens), and for buyers who want published, modellable pricing at the tier above Team. ChatGPT is the stronger choice for broad, lightweight automation across many small workflows, for image generation inside the product, and for sterling billing on its published Business tier. On token price the two are now level at every tier for ordinary-length requests, which was not true a month ago. The decision matrix further down scores each criterion, so you can weight the rows that matter to your business.

What changed in the last 12 months

The model families both turned over, and then turned over again. On the Anthropic side the current line-up is Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5, the last three released between 22 September and 7 October 2026 (Claude models overview, platform.claude.com, 10 October 2026). On the OpenAI side the flagship API family is now GPT-6: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna (OpenAI API pricing page, 10 October 2026), and OpenAI's plan comparison lists all three among the models on its business plans. Availability and limits still vary by plan, so check the plan you are buying rather than the headline. Our briefings on GPT-6 Astra and the Claude 5.5 family cover what those releases do and do not mean.

The long-context gap closed, and reappeared as a pricing difference. Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 each carry a 1M-token context window with 128K maximum output. GPT-6.1 Sol and GPT-6 Luna carry a 1,050,000-token context window with 128,000 maximum output (OpenAI model pages, 10 October 2026). The difference is in the billing. Anthropic states that its current models other than Haiku 5.5 "include the full 1M token context window at standard pricing", adding that "a 900k-token request is billed at the same per-token rate as a 9k-token request". OpenAI states that prompts with more than 272K input tokens on GPT-6.1 Sol "are priced at 2x input and cache rates and 1.5x output for the full request", and the same rule applies to GPT-6 Luna. Claude Haiku 5.5 is Anthropic's one exception: a prompt over 100,000 tokens pays five times the lower rate. If your work routinely involves very large documents, that is the line to model, not the headline window.

Claude's Enterprise tier moved to a published price and a usage-based model. Claude Enterprise is now listed at US$20 per seat per month billed annually, plus usage billed at API rates, with a self-serve route as well as a sales-assisted one. Anthropic's Help Center is explicit that "the seat fee only covers access to the platform and doesn't include any usage", with "no per-seat usage limits and no included token allowance".

ChatGPT Business replaced the old Team tier and gained seat types. It is now sold as standard and premium seats, for teams of 2 to 200 employees, and ChatGPT Enterprise no longer publishes a seat price or a minimum at all: the pricing page says "Custom pricing" and "Contact sales".

Both vendors expanded data residency, and neither extended it to UK processing. That is covered in full below, because it is the point on which the previous version of this guide was least precise.

The product surfaces are still moving. On 16 September 2026 Anthropic merged Claude Cowork and chat into a single Claude. By 10 October its plan comparison listed Claude Design, Slides and Docs on every plan including Free, Dashboards in beta on paid plans and Motion in beta on Team and Enterprise. OpenAI's Business plan description now lists ChatGPT Work alongside ChatGPT and Codex. A business evaluating either platform this quarter should expect the surface it demos to differ from the surface it rolls out.

Pricing in GBP, with VAT

Anthropic's pricing page is localised: a UK visitor sees sterling. OpenAI also prices in sterling for a UK customer. The figures below are what each vendor showed a UK connection on 10 October 2026, and every seat price is unchanged from 17 September.

Claude (claude.com/pricing, UK render): Free £0. Pro £15 a month on annual billing, £180 billed up front, or £18 billed monthly. Max from £90 a month. Team standard seats £18 a seat a month on annual billing or £22.50 monthly; Team premium seats £90 or £113; the Team plan is "For teams of 2 to 150", and now includes up to US$500 a month in pooled API credits for building on the Claude Platform. Claude Enterprise is shown as "Seat price + usage at API rates", priced at "US$20/seat/month, billed annually", with the note that "usage cost scales with model and task".

ChatGPT (the plan picker on a UK account, and chatgpt.com/pricing): Business standard seats £15 a seat a month billed annually or £18 billed monthly; Business premium seats £75 or £90; the Business plan is for teams of 2 to 200 employees. ChatGPT Enterprise carries no published price: the pricing page says "Custom pricing".

The VAT position differs between the two vendors, and you should not assume it away. Anthropic's UK render carries the line "Prices include 20% VAT". Its Help Center adds, for the Team plan, "Prices shown are for US customers and exclude applicable taxes. Pricing, currency, and tax handling vary by region." OpenAI's public pricing page states no tax basis for the Business plan. In the signed-in plan picker on a UK account on 10 October 2026, the personal plan price carried a line stating the VAT it included, and the Business seat prices carried no tax line at all. So Claude's £18 and ChatGPT's £15 are not necessarily stated on the same basis, and the gap between them could be the whole of the difference or none of it. We report what each vendor publishes and stop there. Confirm the basis on a quote or an invoice, and take the VAT treatment itself to your accountant.

One currency detail matters at Enterprise tier. Anthropic's Help Center states that self-serve Enterprise accepts "USD only", and that multi-currency is supported on the sales-assisted route. A UK business that buys Claude Enterprise self-serve therefore carries the exchange rate itself. For the sterling conversion method, the HMRC rate basis and the full worked arithmetic on Claude's prices, see our guide to Claude pricing in the UK, which is the page that owns that question.

API prices are US dollars on every surface for both vendors, and they have converged. Claude: Fable 5.1 $10 input and $50 output per million tokens; Opus 5.5 $4 and $20; Sonnet 5.5 $2 and $10; Haiku 5.5 $0.10 and $0.50 on prompts up to 100,000 tokens, and $0.50 and $2.50 above that. OpenAI: GPT-6 Astra $10 and $50; GPT-6.1 Sol $2 and $10; GPT-6 Luna $0.10 and $0.50, each rising to twice the input rate and 1.5 times the output rate on prompts above 272K input tokens. For ordinary-length requests the two lists are now level at the frontier, the mid tier and the cheapest tier, and Claude Opus 5.5 at $4 and $20 sits between tiers with no like-for-like OpenAI model at that price. A month ago OpenAI was materially cheaper at the bottom of the range; it no longer is, except on long prompts, where GPT-6 Luna's step comes later and is smaller than Haiku 5.5's.

Total cost of ownership at 10, 25 and 50 seats

These are seat costs only, on annual billing, at the standard seat type, using the sterling prices above. Multiply seats by the monthly price by twelve. The arithmetic is deliberately visible so you can substitute your own seat mix.

Seats	Claude Team standard (£18/seat/month)	ChatGPT Business standard (£15/seat/month)	Difference a year
10	£2,160	£1,800	£360
25	£5,400	£4,500	£900
50	£10,800	£9,000	£1,800

Read that table with the VAT caveat above in mind. Anthropic states its sterling prices include 20% VAT; OpenAI states no basis. If OpenAI's £15 turns out to be exclusive of VAT, the two are level at £18 inclusive and the difference collapses to nothing. That single unknown is worth more than any feature in the comparison at these seat counts, and it takes one question to a salesperson to settle.

Premium seats move the numbers sharply. Claude premium at £90 a seat a month is £1,080 a year per user; ChatGPT premium at £75 is £900. Both vendors let you mix seat types, so the realistic model is a handful of premium seats over a standard base rather than a uniform tier.

Claude Enterprise needs its own arithmetic, because the seat fee buys no usage. Anthropic's stated minimums are 20 seats self-serve and 50 sales-assisted. At 50 seats the seat line is 50 × US$20 × 12 = US$12,000 a year, and every token is billed on top at API rates. To size the usage line you need a token assumption, and the only defensible one is your own. As an illustration of the method rather than a forecast: assume one million input tokens and 200,000 output tokens per seat per month. On Sonnet 5.5 at $2 and $10 per million, that is $2 + $2 = $4 per seat per month, so 50 seats over twelve months adds US$2,400. On Opus 5.5 at $4 and $20, the same volume is $4 + $4 = $8 per seat per month, or US$4,800 a year. Identical usage, twice the bill, decided by a model setting.

That is the reason to run a month on Claude Team with usage logged before committing to Enterprise. Team includes usage in the seat fee, so a pilot there gives you a real token profile at a known cost, and that profile is the input the Enterprise model actually needs. Anyone signing Claude Enterprise without one is guessing at the larger half of the bill.

ChatGPT Enterprise cannot be modelled from published figures at all, because OpenAI publishes none. Its pricing page notes that "credit-based and token-based pricing are available for Enterprise plans", which means the same usage question applies, without the published seat rate to anchor it.

UK data residency and GDPR posture

This is the section most likely to change a decision, and the one where the previous version of this guide was too generous to ChatGPT. The correction is simple: neither vendor will run inference in the United Kingdom.

OpenAI separates storage from processing and publishes a region table. The United Kingdom endpoint, gb.api.openai.com, is listed as Storage Yes, Processing No. The only regions carrying Processing Yes are the United States, Europe (EEA and Switzerland), and the United Arab Emirates. OpenAI states that where a region does not support regional processing, it "may also process and temporarily store Customer Content outside of the Region to deliver the services". The same split applies on the ChatGPT side: the Help Center lists data residency for ChatGPT Enterprise and Edu across ten regions including the United Kingdom, but inference residency across only Europe (EEA and Switzerland), the United States and the United Arab Emirates.

Anthropic's position is unchanged and we have documented it separately. The first-party Claude API offers no UK or EU option; in-region processing runs through AWS Bedrock, where only Ireland and Stockholm are designated in-region and London is not, or Google Cloud Vertex AI's EU multi-region, which does not include the United Kingdom. Our guide to Claude data residency for UK organisations sets out the routes, the premiums and the long list of features that stop working in-region.

So the honest comparison is not "ChatGPT has UK residency and Claude does not". It is that both vendors will store UK data at rest in the UK on the right tier or route, neither will process it here, and for both the nearest genuine in-region processing is Europe. Both charge for it: OpenAI states that "data residency endpoints are charged a 10% uplift for models released on or after March 5, 2026", and Anthropic's partner-cloud regional endpoints carry a 10% premium, with US-only inference on the first-party API priced at 1.1 times standard rates.

Two tier-level details matter at SME scale. First, on OpenAI the ten-region data residency row in the pricing comparison is marked Yes for ChatGPT Enterprise and No for ChatGPT Business, so the tier a 10 to 50 seat firm would actually buy does not carry it by default. A separate Help Center article, still current on 10 October 2026, describes a ChatGPT Business residency option that is "rolling out gradually and is not yet available to all customers", covers storage only, and explicitly "does not include inference residency"; it also states that if you choose a non-US region, "a copy of every prompt and response is also stored in the United States for a limited time for safety, abuse monitoring, and enforcement". Read those two statements together before you write a residency requirement into a contract. Second, on Anthropic, US-only inference is an Enterprise feature, not a Team one.

A Data Processing Addendum is required under UK GDPR for any AI vendor processing personal data, and both vendors publish one. Consumer and free tiers carry no such protection. And it is worth testing the requirement itself before designing around it: UK GDPR regulates restricted transfers rather than mandating UK storage, a distinction that our residency guide covers and that regularly turns a residency problem into a transfer-assessment problem. A data protection impact assessment may be required for customer-facing or HR use cases whichever vendor you choose; our DPIA screening test covers when.

Tooling and integrations

Claude's strength remains depth in agentic and document work. Anthropic states that "Claude Code is included in all paid plans", not only business tiers, and the Model Context Protocol gives Claude a standard way to reach business systems. Team and Enterprise add enterprise search across the organisation, organisation-wide skills deployment and connectors for Google Drive, Gmail, Google Calendar, GitHub, Microsoft 365 and Slack. Claude Design, Slides and Docs are now listed on every plan, and Team and Max plans include monthly API credits for building on the Claude Platform.

ChatGPT's strength remains breadth of surface. Custom GPTs, apps connecting to internal tools, a built-in browser, Codex, record mode, interactive tables and charts, and in-product image generation all ship in the product. Workspace agents and company knowledge are marked available on both Business and Enterprise.

Two old differentiators have expired and should not be repeated. Claude now lists voice mode on every tier including Free, so "no voice in product" is out of date; in-product image generation remains a ChatGPT feature and is not listed on Claude's plan comparison. And the practical view holds: if the work is coding-heavy, document-heavy or genuinely agentic, Claude tends to fit better. If the work is many people doing many small things across many tools, ChatGPT's surface area is broader. Most organisations of any size end up running both.

UK-specific decision factors

Cyber Essentials now reaches your AI workspace, and this is the factor most UK SMEs have not priced in. Cyber Essentials version 3.3, question set Danzell, applies to assessment accounts created after 26 April 2026. IASME, which delivers the scheme with the NCSC, states that multi-factor authentication "is a mandatory requirement for all cloud services where it is available" and that organisations failing to implement it "whether it is free, included, or a paid option, will now automatically fail the assessment". The requirements document adds a definition that closes the obvious escape route: "If your organisation's data or services are hosted on cloud services, these services must be in scope. Cloud services cannot be excluded from scope." A Claude or ChatGPT workspace is a cloud service under that definition. Both vendors offer single sign-on on their business tiers and neither does on individual plans, so a team running business work through personal Pro subscriptions has an in-scope cloud service with no organisation-level control over it. If you hold Cyber Essentials or bid into supply chains that require it, that is a procurement constraint, not an IT preference.

Vendor certifications are now close to level, and one previous claim does not survive checking. Anthropic's trust centre shows SOC 2 Type 2, ISO 27001, ISO 42001, CSA STAR and NIST 800-171 across the Claude API, Claude Team and Claude Enterprise, with HIPAA marked for the API and Enterprise but not for Team, and a UK Cyber Essentials certificate held by Anthropic Ireland Limited. OpenAI's trust portal shows SOC 2 Type 2, SOC 3, ISO 27001, 27017, 27018 and 27701, ISO 42001, CSA STAR, PCI DSS and FedRAMP 20x. OpenAI does not list UK Cyber Essentials, so we no longer say that it holds it. Note also that on OpenAI's own plan comparison the ISO certification row is marked for ChatGPT Enterprise and not for ChatGPT Business.

The public-sector signal is no longer one-sided. Both vendors have signed memoranda of understanding with the Department for Science, Innovation and Technology, OpenAI's published on GOV.UK. Anthropic was additionally selected by DSIT in January 2026 to help build and pilot an AI assistant for GOV.UK services. That is a real distinction, but it is narrower than "government prefers Claude", and DSIT's functions moved to the Cabinet Office in July 2026. Treat it as one input to a bid, not a deciding one.

British English handling remains a configuration question rather than a vendor one. Both platforms respect UK spelling and register when the system prompt or organisation instructions say so, and both drift towards American English without it. Anthropic exposes organisation-wide instructions on Team and Enterprise; setting them once is cheaper than correcting output forever.

What a sceptical CFO will push back on

The seat price is not the price. At Team or Business tier it very nearly is, which is why the ten to fifty seat comparison above comes down to a VAT basis rather than a feature. At Claude Enterprise it is not: the seat fee buys access and nothing else, and the usage line is unbounded until an administrator sets a spend limit. Set those limits on day one rather than in month three.

Benchmark scores do not translate into business outcomes. Both vendors publish comparison tables produced on their own harnesses with their own settings. Neither is independent, and neither measures drafting a client letter, summarising a supplier contract or answering a customer. Pilot on your actual tasks with your actual documents, and judge on those.

The plan tier is necessary and not sufficient. Buying a business tier does not by itself give you a compliant deployment. The admin controls have to be configured, the DPA signed, the usage policy written, the MFA enforced and the residency question answered honestly. The tier makes that work possible; it does not do it.

Watch which controls sit above your tier. On Claude, SCIM, audit logs, the Compliance API, role-based access and custom data retention are Enterprise features and are marked unavailable on Team in the pricing comparison. Anthropic's Team plan Help Center article, by contrast, lists role-based permissioning and domain capture among Team features, so ask which applies before relying on either. On ChatGPT, SCIM, enterprise key management, role-based access, the compliance API logs platform, IP allowlisting and the ISO certifications are Enterprise and are marked unavailable on Business. If your auditor expects audit logs, neither mid-tier plan has them, and that discovery is much cheaper before signature than after.

Lock-in is real on both sides. Custom GPTs and workspace agents on one platform, Projects, Skills and MCP connectors on the other, all accumulate switching cost. Ask the exit question at purchase: what does it cost us to move in eighteen months, and what would we have to rebuild?

A decision matrix for UK buyers

Score each criterion 0 for poor fit, 1 for workable, 2 for strong fit, then apply your own weightings. The column headings name the tier a 10 to 50 seat UK firm would realistically buy, because several rows change at Enterprise.

Criterion	Claude (Team, or Enterprise from 20 seats)	ChatGPT (Business, or Enterprise by quotation)
UK storage at rest on the tier an SME buys	0 on Team and on the first-party API; 2 via AWS Bedrock or Google Cloud Vertex AI at a 10% premium	1: Enterprise covers the UK among ten regions, Business residency is storage-only and still rolling out
UK in-region processing	0. Nearest is Ireland or Stockholm on Bedrock	0. UK endpoint is storage only; nearest is Europe, EEA and Switzerland
Long-document work	2: 1M context on Fable 5.1, Opus 5.5 and Sonnet 5.5, billed at standard rates across the full window	2: 1,050,000 context on GPT-6.1 Sol, with 2x input and 1.5x output above 272K input tokens
Coding and agentic workflows	2: Claude Code on all paid plans, MCP connectors, Skills	1 to 2: Codex, workspace agents, apps connecting to internal tools
Broad casual-user automation across many small workflows	1	2: larger custom-tool ecosystem, more lightweight integrations
Image generation in product	0: not listed on the plan comparison	2: image generation and image generation with thinking
Cheapest token tier for high-volume automation	2: Haiku 5.5 at $0.10 in and $0.50 out on prompts up to 100,000 tokens, five times that above	2: GPT-6 Luna at $0.10 in and $0.50 out, with 2x input and 1.5x output above 272K input tokens
Admin controls included at mid tier	1: SSO, central billing, usage analytics, org-wide skills; no SCIM, audit logs, Compliance API or role-based access	1: SAML SSO, admin console, analytics dashboard, SOC 2; no SCIM, EKM, role-based access or ISO certifications
Billing in sterling	1: Team is priced in GBP, self-serve Enterprise is USD only, multi-currency is sales-assisted	2: Business is priced in GBP; Enterprise terms by quotation
Published, modellable pricing at the tier above	2: US$20 a seat plus usage at API rates, minimums stated	0: no published price, no published minimum
Compliance certification breadth	2: SOC 2 Type 2, ISO 27001, ISO 42001, CSA STAR, NIST 800-171; HIPAA on API and Enterprise; UK Cyber Essentials via Anthropic Ireland Limited	2: SOC 2 Type 2, SOC 3, ISO 27001, 27017, 27018, 27701, ISO 42001, CSA STAR, PCI DSS, FedRAMP 20x; ISO row is Enterprise only
UK public-sector signal	2: DSIT MoU plus selection for the GOV.UK assistant pilot	1: DSIT MoU published on GOV.UK

This is a frame, not a verdict. Most UK firms score Claude higher on document-heavy, coding and regulated-sector criteria and ChatGPT higher on breadth, sterling billing and in-product media, with the outcome decided by which rows carry weight in your business.

Sector-specific recommendations

Professional services: law, accountancy, audit, consulting. Claude remains the stronger default. The 1M-token window at flat pricing suits matter files, contract bundles and audit working papers, and the absence of a long-context surcharge matters when large documents are the normal unit of work rather than the exception. See Claude for UK professional services.

Financial services. Either is workable and the decision usually follows the existing cloud relationship: AWS-centred firms reach Claude through Bedrock, Microsoft-centred firms tend towards ChatGPT Enterprise. Note that neither route delivers UK inference, so a mandate expressed as UK-only processing will need to be re-examined rather than procured around. See Claude for UK financial services.

Healthcare and life sciences. Claude is generally stronger for clinical and research document work, but check the tier: Anthropic's trust centre marks HIPAA for the Claude API and Claude Enterprise and not for Claude Team, and the HIPAA-ready configuration with a BAA is an Enterprise feature. ChatGPT Enterprise is workable for administrative and back-office use. Sector governance is required whichever is chosen.

Logistics and operations. Mixed. ChatGPT tends to win on broad operational automation across many light workflows; Claude wins where the work is contracts, compliance and policy, or where agentic workflows are part of the operational stack.

Software and engineering-heavy teams. Claude is generally the stronger default because Claude Code is included on every paid plan and MCP gives a standard integration path. Many engineering-led businesses run Claude for the product teams and ChatGPT for sales and marketing.

Retail, hospitality and consumer-facing SMEs. ChatGPT is generally the stronger default: image generation and the wider app ecosystem suit content, customer service and high-volume light automation.

Public sector and bidders into public-sector work. Claude carries a modestly stronger signal following the GOV.UK assistant selection, but both vendors hold DSIT memoranda, so the deciding factor is more often the buyer's own security schedule, including Cyber Essentials scope, than the vendor's Whitehall relationships.

The honest reading has not changed: for most UK SMEs this is rarely a question of one platform winning outright. It is a question of which platform leads your top three use cases, with the option to add the second later for specific workflows.

Questions to answer before signing

1. Is the sterling price we have been quoted inclusive or exclusive of VAT, and is the other vendor's quoted on the same basis? At 10 to 50 seats this is worth more than any feature in the comparison.
2. What is our projected total at the end of year two, including usage at API rates if Claude Enterprise is on the table, and including the 10% uplift if we need regional processing?
3. Does our requirement genuinely call for UK processing, or for UK storage with a properly assessed transfer? Neither vendor offers the first; both offer the second on the right tier or route.
4. Which controls does our auditor expect, and do they sit inside the tier we are buying or above it? Audit logs, SCIM and role-based access are above the mid tier on both platforms.
5. Is our AI workspace in scope for Cyber Essentials, and is MFA enforced for every user on it?
6. What is our exit plan if we need to migrate in eighteen to twenty-four months, and what would we have to rebuild?

The AI Consultancy runs procurement reviews for UK SMEs comparing Claude, ChatGPT and a deliberate two-vendor posture. We are Certified in the Claude Partner Network and our team holds the Claude Certified Architect credential, and we implement ChatGPT as well. For Claude specifically, see our Claude implementation and Claude enablement services. For ChatGPT, including Enterprise deployment, API integration and staff rollout, see our ChatGPT implementation consultants service. For a readiness assessment before you commit, see our AI readiness assessment, and for the arithmetic on projected savings, our AI ROI calculator.