Our mid-2026 roundup was about a burst of model launches and the shift from chatbots to agents. Four months on, the story is less about raw capability and more about who owns the tools, how they are priced, and where your data sits. Some of those changes are big enough that a tool you chose in the summer may now come with a different owner, a different bill or a different set of models.
This roundup covers what has changed since July, what it costs in pounds, and what a UK team should actually do about it. If you only have two minutes, skip to What this means for UK teams.
A note on dates and sources: everything below is accurate as of 6 November 2026. We’ve drawn it from the vendors’ own pricing and documentation pages wherever possible, and from news reporting where a vendor page doesn’t cover it. Each section links its sources, and the full list is at the end. GBP figures are approximate conversions from US dollar list prices, excluding VAT, so always check the final price at checkout.
The big one: Cursor now belongs to SpaceX
The most consequential change for developers is a change of owner. SpaceX completed its acquisition of Anysphere, the company behind the AI code editor Cursor, on 14 August 2026, in an all-stock deal valued at around £47 billion (TechSpot). Earlier in the year SpaceX had absorbed Elon Musk’s xAI, so Cursor now sits in the same group as the Grok models.
The practical effect arrived on 12 October, when Cursor stopped being sold as a standalone subscription and moved inside xAI’s “X Subscription”, a single bundle covering Grok, Cursor and X Premium with one shared usage pool. Cursor’s own pricing page now lists these tiers (cursor.com/pricing):
| Tier | Approx. price (GBP) | What matters for developers |
|---|---|---|
| Free | £0 | Limited AI usage |
| Plus | ~£6/mo | Grok models only |
| Premium | ~£24/mo | 4× Plus usage; third-party models available but billed separately |
| Super | ~£80/mo | 15× Plus usage |
| Ultra | ~£160/mo | 50× Plus usage |
| Teams | from ~£24/user | Pooled usage |
Two details matter more than the headline prices. First, the cheapest paid tier only gives you Grok models. Second, third-party models such as Claude and GPT are not included in any tier: Cursor’s page says they are billed separately at the provider’s API list price. Many developers chose Cursor precisely to use Claude inside it, so the real monthly cost for that workflow is now Premium plus pay-as-you-go model usage. Existing subscribers on the old Hobby, Pro and Pro+ plans can stay on them for now, as legacy plans.
What to do: if your team uses Cursor, re-check two things before renewing. One is the total cost with your preferred models included. The other is the privacy and data-processing terms under the new owner: earlier assurances were given by a different company. If either is a problem, GitHub Copilot is the obvious alternative, and we’ve updated our Cursor vs GitHub Copilot comparison to reflect the new pricing.
The model line-ups have moved on
Model names change so often that a comparison written in the spring is now describing older versions. Here is where the three main assistants stand in early November.
Anthropic (Claude). Anthropic’s current line-up, per its own model documentation, is Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, all with a one-million-token context window (Claude models overview). Anthropic now recommends Opus 5.5 as the default for most work, and positions Fable 5.1 as the model for the most demanding reasoning and long-running agent tasks. That is notable given that the Fable line was briefly pulled in the US earlier this year, a story we covered in Why the US banned Claude Fable 5 and Mythos 5. For businesses building on the API, list prices per million tokens are:
| Model | Input (approx. GBP) | Output (approx. GBP) | Anthropic’s suggested use |
|---|---|---|---|
| Claude Fable 5.1 | ~£8 | ~£40 | Demanding reasoning, long-horizon agents |
| Claude Opus 5.5 | ~£3.20 | ~£16 | Default for most agentic and knowledge work |
| Claude Sonnet 5.5 | ~£1.60 | ~£8 | Speed and intelligence balance |
| Claude Haiku 5.5 | from ~£0.08 | from ~£0.40 | High-volume classification and extraction |
In the Claude apps, Fable is available on Pro paid from usage credits, and is included on Max up to half the weekly limit (claude.com/pricing).
OpenAI (ChatGPT). ChatGPT’s plans page now lists the GPT-6 family, including GPT-6.1 Sol and GPT-6 Astra, with the Pro plan’s top reasoning tier “powered by GPT-6 Astra” (chatgpt.com/pricing). We could not confirm an official release date from an OpenAI announcement, so we’re not quoting one. What matters for buyers is that the GPT-5.x models most reviews were written about are now the previous generation. The plans page also notes that the cheap Go tier “may include ads”, something to bear in mind before putting it in front of staff.
Google (Gemini). Google’s biggest change for subscribers was price rather than a new model name: it cut Google AI Ultra by roughly 60% at I/O 2026, to around £80 a month (Engadget). The Plus and Pro tiers are unchanged in structure. See our Gemini review for the full line-up.
The takeaway is the one we gave in summer, only more so: don’t choose a tool because of which model it runs this month. Leadership changes every few weeks. Choose on workflow fit, data handling and price, and expect to revisit the choice.
Subscriptions are splitting into more tiers, and the top end is getting pricier
The second clear trend is pricing. A year ago, most AI assistants had a single “pro” plan at about £16–£20 a month. Now nearly every vendor has added heavy-use tiers above it, and several have moved to usage-based billing.
| Product | What changed | Approx. GBP now |
|---|---|---|
| ChatGPT | Pro now comes in three tiers instead of one | Go ~£6, Plus ~£16, Pro from ~£80 (also ~£160 and ~£400) |
| Claude | The top plan (~£160) is now called Ultra, with no five-hour limit and about £160 a month of API credits | Pro ~£16, Max ~£80, Ultra ~£160, Team from ~£16 per seat (annual) |
| GitHub Copilot | Moved to usage-based “AI credits” on 1 June 2026; a Max tier (~£80) was added | Pro ~£8, Pro+ ~£31, Max ~£80, Business ~£15 per seat |
| Google AI | Ultra price cut | AI Ultra from ~£80 |
| Grammarly | The separate Business plan is gone; Pro now covers individuals and teams, and Enterprise is sold through Superhuman Go | Pro ~£10 (billed annually) |
Sources: ChatGPT pricing and BenchLM for the Pro tiers; Claude pricing; GitHub’s announcement; Grammarly plans.
Two practical consequences follow. First, the entry plans are still the right choice for most people. The new upper tiers are built for developers and analysts running long agent sessions, not for someone drafting emails. Second, usage-based billing makes costs harder to predict. Under Copilot’s AI-credit model, each paid plan includes credits roughly equal to its price, and heavier use draws more. For a team, that means setting budget caps in the admin console rather than assuming a flat per-seat cost.
UK data residency is now an option, with a catch
For UK organisations, the most useful change of the year may be the least publicised. OpenAI’s data residency programme now lists the United Kingdom as a storage region, alongside Europe, the US and several others (OpenAI help centre). It is available to new ChatGPT Enterprise and Edu workspaces and to eligible API customers.
The catch is in the detail. Residency covers where your content is stored at rest. Keeping model processing (inference) in-region is currently only offered for Europe, the US and the UAE. So a UK-resident workspace stores data in the UK, but the model may still process it elsewhere. For many organisations, UK storage is enough to satisfy an internal policy. For others, especially in the public sector or health, the processing location matters too, and EU inference residency may be the better fit. It is also worth noting what is not included: ChatGPT Business (formerly Team) and the consumer plans have no residency option at all.
We cover the wider compliance picture, including training defaults and data processing agreements, in Is ChatGPT GDPR compliant?, which we updated this week with links to OpenAI’s and the ICO’s own guidance.
UK regulation: still no AI Act
Despite repeated promises, the UK still has no comprehensive AI law. Law firm Lewis Silkin noted in September that the UK “remains without comprehensive AI legislation” (Lewis Silkin). A private member’s bill on AI superintelligence is due for its second reading on 13 November, but without government backing it is unlikely to progress far (Mondaq).
In practice, AI use in the UK is still governed by existing law: UK GDPR and the Data Protection Act, overseen by the ICO, plus sector regulators such as the FCA and Ofcom. If your organisation also serves customers in the EU, the EU AI Act may apply to you regardless. Our EU AI Act guide for UK businesses explains when and how.
What this means for UK teams
Most of the changes above lead to the same few actions:
- Re-price your AI subscriptions before renewal. Plan names and inclusions have changed at ChatGPT, Claude, Cursor, Copilot and Grammarly since the summer. If you budgeted in spring, the numbers are probably out of date. Our downloadable comparison of every tool in our directory lists current approximate GBP prices.
- If you use Cursor, run the numbers again. With Claude and GPT now billed on top of the bundle, compare the full monthly cost against GitHub Copilot or direct API use, and re-read the data terms under the new owner.
- Use the entry tier unless you have evidence you need more. The new top tiers are aimed at long agent sessions. Start on Plus or Pro, watch your usage, and upgrade only when people genuinely hit limits.
- Set spend caps where billing is usage-based. Copilot’s AI credits and Cursor’s third-party model billing can both grow quietly. Set admin budgets and alerts.
- If data location matters, ask about UK or EU residency. OpenAI’s UK storage option is real progress, but check whether you also need in-region processing.
- Don’t wait for a UK AI Act. Your obligations already exist under UK GDPR. Run a DPIA for significant new AI use, keep a written staff policy, and keep personal data out of consumer accounts.
Frequently asked questions
Is Cursor still worth it after the SpaceX deal? For developers who value whole-codebase editing, the product itself hasn’t been made worse by the change of owner. What has changed is the price structure and who holds your data. If you mainly use Grok models, the new bundle may even be cheaper. If you rely on Claude or GPT inside Cursor, expect to pay more than before, and compare it with GitHub Copilot.
Which Claude model should a business use now? Anthropic’s own guidance is to start with Claude Opus 5.5 for most work, and to move up to Fable 5.1 only for the most demanding reasoning or long-running agent tasks, where its higher price is justified.
Do I need ChatGPT Pro? Probably not. Plus at about £16 a month covers most professional use. Pro, from about £80 a month, is aimed at heavy reasoning, long research and coding sessions.
Can I keep ChatGPT data in the UK? On new ChatGPT Enterprise and Edu workspaces and eligible API accounts, yes for storage at rest. Processing may still happen outside the UK, and Business and consumer plans have no residency option.
Is there a UK AI law I need to comply with? Not a dedicated one yet. Existing data protection law applies in full, and the EU AI Act can apply if you serve EU customers.
Sources
- SpaceX acquisition of Cursor: TechSpot
- Cursor pricing and X Subscription tiers: cursor.com/pricing
- Claude models and API prices: Claude models overview
- Claude plans: claude.com/pricing
- ChatGPT plans and models: chatgpt.com/pricing; Pro tier prices: BenchLM
- GitHub Copilot billing change: GitHub blog
- Google AI Ultra price cut: Engadget
- Grammarly plans: grammarly.com/plans
- OpenAI data residency regions: OpenAI help centre
- UK AI legislation status: Lewis Silkin; Mondaq
Final thoughts
The summer was about what AI models could do. The autumn has been about the business around them: acquisitions, bundles, tiered pricing and data location. None of it changes the basic advice. Pick tools for the job, keep personal data on business plans with a data processing agreement, and review your choices every few months. What has changed is how quickly a sensible choice can go stale. If you set up your AI stack before the summer, now is a good time to check it again.
This article is general information, not legal or financial advice. Prices are approximate GBP conversions of US list prices and change often; confirm them with the vendor. See our editorial standards for how we research and update content.