LLM Cost in the UK: What You'll Actually Pay in 2025
If you're a UK-based developer or business trying to figure out what LLMs actually cost, the short answer is: it depends on the model, but you'll almost always pay more than the headline USD price suggests once currency conversion, VAT, and provider fees are factored in. Most LLM providers price in US dollars, which means UK teams are exposed to exchange rate movement on top of the base token cost.
This article breaks down the real cost components UK teams need to budget for, gives current pricing context, and covers the practical steps to keep spend predictable when your invoices are denominated in a currency you don't operate in.
Why LLM cost in the UK isn't just the USD price
Nearly every major LLM provider — OpenAI, Anthropic, Google — bills in USD by default. For a UK business, that introduces three extra cost layers beyond the per-token price:
- Currency conversion. Your card issuer or payment processor applies an exchange rate at the time of billing, plus often a foreign transaction fee of 1–3%.
- VAT. UK VAT (currently 20%) generally applies to digital services purchased by UK businesses and consumers, even when the vendor is based in the US. Depending on your VAT registration status, this may be reclaimable or a genuine added cost.
- Exchange rate volatility. GBP/USD moves month to month. A model that cost you £X in January can cost noticeably more or less in June with zero change in your usage pattern.
None of this shows up when you look at a provider's pricing page, which is why actual invoices tend to surprise finance teams more than engineering teams.
What LLM APIs actually cost per token
Pricing varies by model tier and changes periodically, but as a rough framework:
- Small/fast models (e.g. Claude Haiku-class, GPT-4o-mini-class): typically fractions of a cent to a few cents per 1,000 tokens.
- Mid-tier models (e.g. Claude Sonnet-class, GPT-4o-class): a few cents per 1,000 tokens, with output tokens priced higher than input tokens.
- Frontier models (e.g. Claude Opus-class, GPT-4-class top tier): significantly more per 1,000 tokens, often 5–10x the mid-tier rate.
The split between input and output pricing matters more than people expect. Output tokens (what the model generates) are usually 3–5x more expensive than input tokens (what you send it). A chatbot that generates long responses will cost meaningfully more than one that mostly summarizes short inputs, even at identical token counts.
A rough monthly estimate
For a small UK SaaS product doing:
- 50,000 requests/month
- ~500 input tokens and ~300 output tokens per request average
- A mid-tier model
That's roughly 25M input tokens and 15M output tokens monthly. Depending on the model, that lands somewhere between £150 and £600/month before VAT and currency conversion — the range is wide because output-heavy use cases and frontier models push costs up quickly.
Fixed vs usage-based access: the other UK-specific consideration
Some teams already pay for a Claude subscription (Pro or Team) for internal use — writing, research, coding assistance — and separately need programmatic API access for a product feature. Running both means paying twice: once for the subscription, once for metered API usage, in two different billing relationships, both in USD.
This is where a flat-fee approach changes the calculation. SubToAPI turns an existing Claude subscription into an HTTPS API with a fixed monthly price in EUR rather than metered USD billing — Solo at €9/month, Team at €19/seat, Scale at €49/seat. For a UK team, that means one predictable invoice instead of a variable USD bill that fluctuates with usage and exchange rates. You get application API keys (sub_live_...), streaming, tool use, and usage metadata through a standard API, without a second metered account to reconcile. There's a free trial at signup if you want to compare it against your current usage-based bill for a month.
It's not the right fit for every workload — extremely high-volume production traffic with unpredictable spikes may still be cheaper on pure metered pricing at scale — but for teams whose usage is steady and who already pay for Claude access, it removes both the token-cost variance and the currency conversion sitting on top of it.
How to keep LLM spend predictable in GBP
- Set a budget in GBP, not USD. Convert your expected monthly token spend to GBP using a buffer of 5–10% for exchange rate movement, not the spot rate on the day you build your estimate.
- Track input/output token ratios separately. Output-heavy features (long-form generation, chat) cost disproportionately more than input-heavy ones (classification, extraction, summarization).
- Check VAT treatment with your accountant. Whether VAT is reclaimable depends on your registration status and whether the invoice is treated as a reverse-charge supply.
- Consider fixed-fee options for steady workloads. If your usage is consistent month to month, a flat subscription-based API can be cheaper and easier to forecast than metered billing — see /pricing for a concrete comparison.
- Cache and reuse where possible. Repeated system prompts, RAG context, or few-shot examples all cost tokens every single call unless you're using prompt caching where the provider supports it.
If you're evaluating a subscription-based approach, the quickstart guide walks through getting an API key and making your first request in a few minutes, and the docs cover streaming and tool use for anyone migrating existing metered API code.
Questions
Does UK VAT apply to LLM API subscriptions from US providers? Generally yes — digital services sold to UK businesses and consumers by non-UK vendors typically fall under UK VAT rules, often via reverse charge for VAT-registered businesses. Confirm treatment with your accountant, as it affects your net cost.
Is it cheaper to use a fixed-fee API or pay per token? It depends on volume and consistency. Steady, moderate usage often costs less with a fixed monthly fee like SubToAPI's plans; highly variable or very high-volume traffic can be cheaper on metered pricing, especially with volume discounts.
Why do LLM costs in the UK feel higher than the US price suggests? Because USD pricing doesn't include currency conversion fees, exchange rate movement, or UK VAT — all of which add to the final invoice a UK business actually pays, even though the provider's advertised per-token rate hasn't changed.