Which Claude model should most UK businesses use?

Claude Sonnet 5 for the majority of production work, Claude Haiku 4.5 for high-volume classification and extraction, and Claude Opus 5 reserved for the tasks where a wrong answer is expensive. Sonnet 5 sits at $2 per million input tokens and $10 per million output tokens, which is a third less than the Claude Sonnet 4.6 it succeeds, and it carries the same 1M token context window and 128k maximum output as Opus 5. The usual mistake is not picking the wrong model, it is picking one model for everything. Routing by task is worth doing on the published prices alone: Opus 5 input costs two and a half times Sonnet 5 and five times Haiku 4.5, so any workload moved down a tier saves that multiple on the tokens it consumes. Whether quality holds at the lower tier is a question to settle by measuring your own outputs, not by taking a vendor's or a consultancy's word for it.

How much does Claude Sonnet 5 cost in pounds per million tokens?

At HMRC's published monthly exchange rate for August 2026 of USD 1.3367 to GBP 1, Claude Sonnet 5's $2 input and $10 output per million tokens works out at roughly £1.50 and £7.48. Through the Batch API at half price that becomes about £0.75 and £3.74. Anthropic publishes API prices in US dollars only and bills in dollars, so any sterling figure is a dated conversion rather than a published price, and it moves when HMRC publishes a new monthly rate.

Is Claude Sonnet 5's introductory pricing about to end?

No. It launched on 30 June 2026 with $2 and $10 per million tokens described as introductory pricing through 31 August 2026, with a rise to $3 and $15 due on 1 September. Anthropic confirmed in an update dated 10 August 2026 to that announcement that the introductory rates are now the standard price and the September increase will not occur. If you see $3 and $15 on an Anthropic pricing page, check which row it is on: that is the current price of Claude Sonnet 4.6 and Sonnet 4.5, not a pending Sonnet 5 change.

Does using the 1M token context window cost more per token?

No. Anthropic states that Claude 4.6 and later models include the full 1M token context window at standard pricing, and gives the explicit illustration that a 900k-token request is billed at the same per-token rate as a 9k-token request. Prompt caching and batch discounts also apply at standard rates across the full window. The cost of a large context is simply that it contains more tokens, which is a reason to be deliberate about what you put in the window rather than a reason to avoid long-context models. Note that this no-premium statement covers Claude 4.6 and later; do not assume it applies to earlier models.

What actually reduces a Claude API bill?

Four levers, usually in this order. The Batch API takes 50 per cent off both input and output tokens for anything that does not need a synchronous answer. Prompt caching charges cache reads at a tenth of the input rate, with writes at 1.25x for the five-minute cache or 2x for the one-hour cache, so Anthropic's own break-even is one read at five minutes or two reads at an hour. The effort parameter reduces thinking volume and tool calls at lower settings. And tool-use system prompt overhead is real: on Claude Opus 5 the tool system prompt costs 286 tokens against 675 on Opus 4.7, so a tool-heavy workload gets cheaper simply by moving to a newer model. Batch and caching discounts can be combined.

Can a UK business buy UK data residency for Claude?

Not as a published option on the first-party API. Anthropic documents two values for the inference_geo parameter on Claude 4.6 and later models: global, which is the default at standard pricing, and us, which pins inference to United States capacity at a 1.1x multiplier on all token categories. There is no UK or EU residency tier with a published price. On the partner clouds, regional and multi-region endpoints carry a 10 per cent premium over global endpoints. Residency and cost need deciding together rather than in sequence, and our guide to Claude data residency for UK organisations sets out what each route actually offers.

Back to Blog

strategy

Which Claude model should a UK business use in 2026?

By Jay MatharuPublished 12 August 2026Last reviewed 12 August 2026

Two colleagues walking and talking along a glazed office corridor, with the City of London skyline soft beyond the glass

Anthropic released four Claude 5 models between 9 June and 24 July 2026. For a UK business trying to make one decision and then get on with the work, that pace is the problem: any model comparison written more than a few weeks ago is describing a lineup that no longer exists, and the instinct to standardise on the most capable model available is the most expensive instinct in the room.

This guide takes the opposite approach. It starts from the workload, not the model, and it treats the price list as one input among several. Every figure was verified against Anthropic's own documentation on 12 August 2026, and the sterling conversions state their basis, because a Claude cost in pounds is only as good as the exchange assumption behind it.

The current lineup, and what each model is actually for

Five models matter to a UK business today. Four are generally available and one is not.

Model	USD per million tokens (in / out)	Context / max output	Best fit
Claude Haiku 4.5	$1 / $5	200k / 64k	High volume, low ambiguity: classification, extraction, routing, first-pass triage
Claude Sonnet 5	$2 / $10	1M / 128k	The default for production work: drafting, summarising, document review, most agentic tasks
Claude Opus 5	$5 / $25	1M / 128k	Tasks where a wrong answer is expensive: complex reasoning, difficult code, long agentic runs
Claude Fable 5	$10 / $50	1M / 128k	The frontier tier, with adaptive thinking always on
Claude Mythos 5	$10 / $50	Shares Fable 5's specs	Not generally available: see the note below

Prices correct as at 12 August 2026. Anthropic has shipped four Claude 5 models in under two months, so treat any price table in any article, including this one, as a dated snapshot rather than a standing fact, and check the current figures before committing a budget.

Two things in that table are worth pausing on. First, Claude Sonnet 5 is cheaper than the model it replaced. Claude Sonnet 4.6 and Sonnet 4.5 are both priced at $3 input and $15 output per million tokens, so moving up to Sonnet 5 cuts the per-token price by a third. That is an unusual direction of travel and it changes the arithmetic of a mixed deployment: the gap between Haiku and Sonnet is now narrower than the gap between Sonnet and Opus.

Second, the context window is no longer a differentiator between the mid and top tiers. Fable 5, Opus 5 and Sonnet 5 all carry a 1M token window and a 128k maximum output. If you were reaching for Opus purely to fit a large document set into a single call, that reason has gone. One caveat that catches people out: those are the API model specifications. The subscription plans list smaller windows, 200k tokens on Free, Pro, Max and Team and 500k on the Enterprise tiers' default model, so the window you get depends on the route as well as the model.

On Claude Mythos 5, the accurate position is narrow. Anthropic offers it in limited availability to approved customers in Project Glasswing, for defensive cybersecurity workflows, and states that access is invitation-only with no self-serve sign-up. Its launch announcement describes it as restricted to Glasswing partners with cyber safeguards lifted, and soon to select biology researchers, until a broader trusted access programme is available. Prospective users are told to contact their Anthropic, AWS or Google Cloud account team. That is the whole of what is published, and a UK buyer should plan on the basis that Mythos 5 is not part of the available lineup.

Claude Sonnet 5's pricing is now permanent, and that matters more than it sounds

Claude Sonnet 5 launched on 30 June 2026 with $2 and $10 per million tokens presented as introductory pricing running through 31 August 2026, after which the standard rate of $3 and $15 was due to apply from 1 September. Anthropic confirmed in an update dated 10 August 2026 to that launch announcement that the introductory pricing is now the standard price, and that the previously scheduled increase will not occur.

For anyone building a business case, the significance is not the two dollars. It is that the cheapest genuinely capable production model on the list no longer has an expiry date attached to it. A unit-cost model built on Sonnet 5 in July needed a September column; the same model built today does not.

One anti-error note, because it will come up. The figures $3 and $15 still appear on Anthropic's pricing pages, and they still appear in the cost-performance charts inside the Sonnet 5 announcement. On the pricing page they are the current price of Sonnet 4.6 and Sonnet 4.5. In the charts they are the superseded numbers, and Anthropic has captioned them to say the introductory pricing has since been made permanent so the actual cost is lower than shown. Neither is a pending Sonnet 5 increase. If someone brings you a screenshot, check the model row.

The same prices in sterling

Anthropic publishes API prices in US dollars and bills in dollars, so a sterling figure has to be derived. These conversions use HMRC's published monthly exchange rate for August 2026, USD 1.3367 to GBP 1, published on 22 July 2026 and valid for the calendar month. We use HMRC's rate rather than a spot rate because it is official, dated and fixed for a known period, which is what a budget needs.

Model	GBP per million tokens (in / out)	Same workload through the Batch API
Claude Haiku 4.5	£0.75 / £3.74	£0.37 / £1.87
Claude Sonnet 5	£1.50 / £7.48	£0.75 / £3.74
Claude Opus 5	£3.74 / £18.70	£1.87 / £9.35
Claude Fable 5	£7.48 / £37.41	£3.74 / £18.70

These figures exclude tax and are indicative rather than invoiced amounts: API billing is in dollars, so what lands on a card statement also depends on the issuer's own rate and any non-sterling fee. Our sterling worked example for Claude pricing sets out the full method, including the subscription side, where Anthropic does publish prices in pounds to UK visitors.

Choosing by workload: five patterns and the model each one wants

The useful unit of decision is a workload, not an organisation. Most UK deployments contain three or four distinct workloads with genuinely different requirements, and the cost difference between matching them properly and standardising on one model is large.

High volume, low ambiguity. Classifying inbound email, extracting fields from invoices, routing enquiries, tagging documents. Haiku 4.5 at $1 and $5 is the right tool and the 200k context window is rarely the binding constraint. If the task is genuinely mechanical, the quality difference against Sonnet is small and the cost difference is a factor of two on input.

Drafting and summarising with a human reviewer. Client correspondence, meeting notes, first drafts of reports. Sonnet 5 is the default. A human is reading the output before it goes anywhere, so the marginal value of a more capable model is largely absorbed by the review step you were going to do anyway.

Document review across a large corpus. Here the 1M token window earns its place, and since it carries no pricing premium on Claude 4.6 and later, the question is only whether the reasoning is hard enough to justify Opus. Start on Sonnet 5, measure, and escalate the cases that fail rather than the whole workload.

Work where a wrong answer is expensive. Complex analysis, difficult refactors, anything feeding a decision that is hard to reverse. Opus 5 at $5 and $25 is the honest choice, and the sensible framing for a finance conversation is not the token price but the cost of the error being avoided.

Long agentic runs with many tool calls. Model choice matters here for a reason that is not on the price list: tool-use overhead. Anthropic publishes a per-model system prompt cost for tool use, and the spread is wide. On Opus 5 a tool choice of auto or none costs 286 tokens against 675 on Opus 4.7, with 406 against 804 for a forced tool choice. A tool-heavy workload therefore gets cheaper by moving to a newer model even at the same headline rate.

The levers that move a bill more than the model row does

In our experience the model choice is usually the third or fourth largest determinant of a Claude bill. These are the levers that tend to matter more.

Batch processing. The Batch API applies a 50 per cent discount to both input and output tokens in exchange for asynchronous turnaround. Anything overnight, anything scheduled, anything where nobody is waiting: this is the single largest available saving and it requires no change to the prompt.

Prompt caching. Cache reads cost 0.1x the base input rate. Writes cost 1.25x for the five-minute cache or 2x for the one-hour cache. Anthropic states the break-even explicitly: caching pays off after one cache read at the five-minute duration, or two reads at the one-hour duration. For any workload with a stable system prompt or a repeated document, this is close to free money. The two discounts stack: batch and caching can be combined.

The effort parameter. Effort takes five values (max, xhigh, high, medium, low), is set as output_config.effort, needs no beta header, and defaults to high, so setting high is identical to omitting it. Lower effort means less thinking and fewer tool calls. Three cautions, all from Anthropic's own documentation. Effort is a behavioural signal rather than a strict token budget. On Opus 5 it controls thinking volume and does not reliably shorten visible responses, so prompt for length if that is what you want. And changing effort between requests invalidates the prompt cache, because effort shapes the rendered prompt, so hold it constant within a cached conversation. Two levers that fight each other is exactly the sort of detail that turns a projected saving into a surprise.

Model support is not uniform. Effort is supported on Fable 5, Mythos 5, Opus 5, Opus 4.8, Opus 4.7, Opus 4.6, Sonnet 5 and Sonnet 4.6, among others. Haiku 4.5 is not listed among the supported models, so effort is not a lever on your cheapest tier. The narrower xhigh level is available on Fable 5, Mythos 5, Opus 5, Opus 4.8, Opus 4.7 and Sonnet 5, while max additionally covers Opus 4.6 and Sonnet 4.6. On Opus 5, requests that try to disable thinking at xhigh or max effort return a 400 error.

Fast mode, and when not to buy it. Fast mode is a research preview that delivers up to 2.5 times higher output tokens per second, supported only on Opus 5 and Opus 4.8, at $10 input and $50 output per million tokens. That is double the standard Opus 5 rate and identical to Fable 5's standard rate, which is the comparison worth making before enabling it. Anthropic is explicit that fast mode runs the same model weights with no change to intelligence or capabilities, that the benefit is in output tokens per second rather than time to first token, and that access is gated through an account manager. It is unavailable on the Batch API, on the partner clouds, and with a Priority Tier commitment, and switching between fast and standard speed invalidates the prompt cache. Buy it for a latency requirement you can name, not for general speed.

The tokenizer change, and why you cannot do this on a calculator

Anthropic's documentation states that Claude 4.7 and later models use a newer tokenizer which produces approximately 30 per cent more tokens for the same text, and that Sonnet 4.6 and earlier use the previous tokenizer. The models overview page corroborates it independently, and you can see it in the context-window tooltips: 1M tokens is described as roughly 750,000 words on Opus 4.6 and Sonnet 4.6, and roughly 555,000 words on Fable 5, Opus 5 and Sonnet 5.

The trap is treating 30 per cent as a conversion factor. Anthropic's own wording is that the exact increase depends on the content and workload shape, and its advice is to re-measure with the token counting endpoint. For the Sonnet 4.6 to Sonnet 5 move specifically, Anthropic's launch post does put a bound on it, giving a content-dependent range of roughly 1.0 to 1.35 times. So that move buys a one-third price cut against a volume increase somewhere between nothing and about a third, and the net effect on the bill is a measurement question rather than an arithmetic one. Anyone who multiplies the two percentages together has produced a number no source supports.

Two second-order effects catch engineering teams rather than finance teams. If the same text occupies more tokens, a context window holds less of your document than it did, and an output ceiling tuned on the previous generation may truncate an equivalent response. Neither announces itself as a cost problem, and both become one when somebody raises the limits to fix them.

Shelf life: how long is this decision good for?

Better than the release cadence suggests. As at 12 August 2026 no model in the current lineup is deprecated, and Anthropic commits to at least 60 days' notice before retiring a publicly released model, notifying customers with active deployments by email and in the documentation. The published tentative retirement dates are not sooner than 24 July 2027 for Opus 5, 30 June 2027 for Sonnet 5, 9 June 2027 for Fable 5, and 15 October 2026 for the Haiku 4.5 snapshot, which is the nearest horizon of the four and worth a diary note.

Two caveats belong in a procurement conversation. Retirement dates published by Anthropic apply to Anthropic-operated platforms; Amazon Bedrock and Google Cloud set their own schedules, so a model's lifecycle status can differ there. And a genuine breaking change has already landed: on Claude 4.7 and later models, setting temperature, top_p or top_k to a non-default value returns a 400 error. Anthropic's recommended replacement is to omit them and use prompting instead. That is an API-level change, not guidance, and it is the sort of thing that turns a model upgrade into an afternoon of debugging.

Where residency and cost meet

For UK organisations with data-location requirements, the position on the first-party API is narrow and should not be overstated. Anthropic documents two values for inference_geo on Claude 4.6 and later models: global, the default, at standard pricing, and us, which pins inference to United States capacity at a 1.1x multiplier applied to input, output, cache writes and cache reads alike. No UK or EU residency tier is documented with a published price. On the partner clouds, regional and multi-region endpoints carry a 10 per cent premium over global endpoints, for Sonnet 4.5, Haiku 4.5, Opus 4.5 and all subsequent models. Earlier models reject the parameter with a 400 error.

The practical consequence is that residency constraints can push a UK deployment onto a route with different economics and a different retirement schedule, which is why the two decisions should be taken together. Our guide to Claude data residency for UK organisations covers what each route offers, and if your rollout involves personal data, the DPIA screening test is the right next step before switching anything on.

A short method you can run this week

List your workloads, not your teams. Assign each one a default of Sonnet 5. Move the mechanical, high-volume ones down to Haiku 4.5 and measure whether quality holds. Move up to Opus 5 only where you can name the cost of a wrong answer. Then, before anyone argues about model choice again, turn on batching for everything asynchronous and caching for every stable prefix, and re-measure your token counts on the current tokenizer rather than trusting last quarter's numbers. Most of the saving is in those last two sentences.

Where The AI Consultancy fits

Choosing a model per workload, measuring where the tokens actually go, and building a unit-cost model that survives the next release is the kind of work our Claude implementation engagements cover, with the advisory side handled through Claude consulting. If you are choosing between plan routes rather than models, our Claude Team versus Enterprise guide covers the feature split, and the Opus 5 migration briefing covers what changed at the top of the range.

All prices and model specifications verified on 12 August 2026 against Anthropic's published pricing documentation, models overview, model deprecations page, effort and fast mode documentation, and the launch announcements for Claude Opus 5 (24 July 2026), Claude Sonnet 5 (30 June 2026), Claude Fable 5 and Claude Mythos 5 (9 June 2026) and Claude Haiku 4.5 (15 October 2025). Sterling conversions use HMRC's monthly exchange rate for August 2026, USD 1.3367 to GBP 1, published 22 July 2026. Prices exclude tax. Anthropic publishes API prices in US dollars only, so all sterling figures are indicative conversions. This is general information, not procurement or tax advice.

Frequently asked questions

Which Claude model should most UK businesses use?: Claude Sonnet 5 for the majority of production work, Claude Haiku 4.5 for high-volume classification and extraction, and Claude Opus 5 reserved for the tasks where a wrong answer is expensive. Sonnet 5 sits at $2 per million input tokens and $10 per million output tokens, which is a third less than the Claude Sonnet 4.6 it succeeds, and it carries the same 1M token context window and 128k maximum output as Opus 5. The usual mistake is not picking the wrong model, it is picking one model for everything. Routing by task is worth doing on the published prices alone: Opus 5 input costs two and a half times Sonnet 5 and five times Haiku 4.5, so any workload moved down a tier saves that multiple on the tokens it consumes. Whether quality holds at the lower tier is a question to settle by measuring your own outputs, not by taking a vendor's or a consultancy's word for it.
How much does Claude Sonnet 5 cost in pounds per million tokens?: At HMRC's published monthly exchange rate for August 2026 of USD 1.3367 to GBP 1, Claude Sonnet 5's $2 input and $10 output per million tokens works out at roughly £1.50 and £7.48. Through the Batch API at half price that becomes about £0.75 and £3.74. Anthropic publishes API prices in US dollars only and bills in dollars, so any sterling figure is a dated conversion rather than a published price, and it moves when HMRC publishes a new monthly rate.
Is Claude Sonnet 5's introductory pricing about to end?: No. It launched on 30 June 2026 with $2 and $10 per million tokens described as introductory pricing through 31 August 2026, with a rise to $3 and $15 due on 1 September. Anthropic confirmed in an update dated 10 August 2026 to that announcement that the introductory rates are now the standard price and the September increase will not occur. If you see $3 and $15 on an Anthropic pricing page, check which row it is on: that is the current price of Claude Sonnet 4.6 and Sonnet 4.5, not a pending Sonnet 5 change.
Does using the 1M token context window cost more per token?: No. Anthropic states that Claude 4.6 and later models include the full 1M token context window at standard pricing, and gives the explicit illustration that a 900k-token request is billed at the same per-token rate as a 9k-token request. Prompt caching and batch discounts also apply at standard rates across the full window. The cost of a large context is simply that it contains more tokens, which is a reason to be deliberate about what you put in the window rather than a reason to avoid long-context models. Note that this no-premium statement covers Claude 4.6 and later; do not assume it applies to earlier models.
What actually reduces a Claude API bill?: Four levers, usually in this order. The Batch API takes 50 per cent off both input and output tokens for anything that does not need a synchronous answer. Prompt caching charges cache reads at a tenth of the input rate, with writes at 1.25x for the five-minute cache or 2x for the one-hour cache, so Anthropic's own break-even is one read at five minutes or two reads at an hour. The effort parameter reduces thinking volume and tool calls at lower settings. And tool-use system prompt overhead is real: on Claude Opus 5 the tool system prompt costs 286 tokens against 675 on Opus 4.7, so a tool-heavy workload gets cheaper simply by moving to a newer model. Batch and caching discounts can be combined.
Can a UK business buy UK data residency for Claude?: Not as a published option on the first-party API. Anthropic documents two values for the inference_geo parameter on Claude 4.6 and later models: global, which is the default at standard pricing, and us, which pins inference to United States capacity at a 1.1x multiplier on all token categories. There is no UK or EU residency tier with a published price. On the partner clouds, regional and multi-region endpoints carry a 10 per cent premium over global endpoints. Residency and cost need deciding together rather than in sequence, and our guide to Claude data residency for UK organisations sets out what each route actually offers.

Get new briefings by email

The AI Consultancy newsletter delivers briefings and analysis for UK businesses. We use your address only to send it, and you can unsubscribe at any time.