The Claude 5.5 family is complete: cheaper at every tier, and four things to check before you switch

What changed

Between 22 September and 7 October 2026 Anthropic replaced its whole mainstream line-up. Claude Opus 5.5 arrived on 22 September at $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5, with cache reads cut from $0.50 to $0.20. Claude Sonnet 5.5 followed on 28 September at Sonnet 5's price of $2 and $10. Claude Haiku 5.5 completed the family on 7 October at $0.10 and $0.50 for prompts up to 100,000 tokens and $0.50 and $2.50 above that, against $1 and $5 for Haiku 4.5. On the same day Anthropic halved Sonnet 5.5's cache-read price from $0.20 to $0.10. All three are available on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure.

Anthropic attaches a saving to each model: Opus 5.5 costs 40% less than Opus 5 on typical workloads at default settings, Sonnet 5.5 up to 30% less per task than Sonnet 5, and Haiku 5.5 around 75% less on average than Haiku 4.5. Those are the vendor's estimates from its own testing, and they combine the list price with a claim that the new models use fewer tokens for the same work.

Two plan changes came with the models. On 22 September Anthropic raised five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans. From the week of 7 October, Max and Team subscribers receive a monthly API credit for the Claude Platform. On Team it is $20 per Standard seat and $100 per Premium seat, pooled into one balance and capped at $500 a month. Enterprise, Pro and Free plans are not eligible.

Why it matters for UK business

A price list that moves in the buyer's favour at every tier is unusual, and it is real: the list prices are published and fixed. The percentages are a different kind of number. Whether a given bill falls by 40%, 15% or not at all depends on how much of the traffic is cached context, how long the prompts are and how the model is configured. Budget from the list prices and your own token counts.

The point most coverage missed is about retention, not price. Anthropic says Opus 5.5 performs at the level of Claude Fable 5.1 on most work. Fable 5.1 is a Covered Model, which means prompts and outputs are retained for at least 30 days by default on every surface. Opus 5.5 is not on the Covered Models list as we read it on 10 October 2026, and Anthropic's launch post states that it is available with zero data retention, as is Sonnet 5.5. For UK firms in regulated or client-confidential work that kept Fable out of their estate in June and have been waiting for Enterprise Frontier Safeguards, this is the nearer route: capability Anthropic describes as close to the frontier, inside a retention position the firm already holds. The cautions are that zero data retention is an agreement rather than a default, that Anthropic says the list will change as designations change, and that "most work" is the vendor's phrase. Anthropic itself notes that benchmark margins at this level have become a less reliable guide to real differences.

Four details change how the new models behave in production.

• Haiku 5.5 is priced by prompt size. A prompt above 100,000 tokens costs five times the lower rate. Anthropic says about 90% of requests to the previous Haiku fell under that line. A workload that feeds long documents to Haiku is the other 10%, and although it still pays half the Haiku 4.5 rate, it will not see the headline saving.
• The tokeniser changed. Haiku 5.5 uses an updated tokeniser, similar to Sonnet 5.5's and Opus 5.5's, and Anthropic says it uses slightly more tokens per task. Compare cost per task, not cost per token.
• Thinking can no longer be turned off on Opus 5.5. Sonnet 5.5 integrations that run with thinking off must move to a new between_tools setting before migrating. Both are documented in Anthropic's migration guides, and both will break an integration that assumed the old behaviour.
• Cyber safeguards now reach Opus and Sonnet. On Opus 5.5 most cybersecurity tasks are rerouted to Opus 4.8, and on Sonnet 5.5 higher-risk ones fall back to Sonnet 5. Routine bug finding and fixing is unaffected. Security teams doing more than that will need to apply to Anthropic's Cyber Verification programme.

The Team API credit is easy to misread. It covers the Claude Platform only: the Claude API, Claude Managed Agents and the Claude Agent SDK. It does not cover interactive Claude Code, extra usage in Claude or Cowork, or Claude on Bedrock, Vertex AI or Foundry. It does not roll over. One owner claims it by linking a Claude Console organisation, which cannot then be changed without contacting support, and everyone with an API key in that organisation draws on the same pool. A team of five Standard seats receives $100 a month. It is a prototyping allowance, not a discount on the seats.

What to do, and what not to do

Do:

• Re-run your cost model with the new list prices and your own token counts, including your cached share. Quote that result, not a vendor percentage.
• If Fable was excluded on retention grounds, evaluate Opus 5.5 under your existing retention settings, and record the model, the position and the date in the DPIA.
• Read the migration guide before moving. Test Opus 5.5 with thinking on, and change the Sonnet thinking setting first if you run it off.
• Measure what share of your Haiku prompts exceed 100,000 tokens before moving batch workloads.
• If you own a Team plan, claim the credit into the Console organisation you intend to build in, and set workspace spend limits so one project cannot drain the pool.

Do not:

• Put 40%, 30% or 75% into a budget. They are estimates on Anthropic's own workloads.
• Assume zero data retention applies to you because the model supports it. It applies if you hold the agreement.
• Treat Opus 5.5 and Fable 5.1 as interchangeable for every task. Test on your own hardest work.

Where The AI Consultancy fits

Choosing the tier, settling the retention position and migrating without breaking an integration is the core of our Claude implementation service. Our guide to choosing a Claude model works through the 5.5 line-up workload by workload, the sterling figures are in what Claude costs in the UK, plan features are in Claude Team versus Enterprise, and the Fable retention background is in our Fable 5.1 briefing.

Verified on 10 October 2026 against Anthropic's launch posts for Claude Opus 5.5 (22 September 2026), Claude Sonnet 5.5 (28 September 2026) and Claude Haiku 5.5 (7 October 2026), and the Claude Help Center articles "Monthly API credits for Max and Team plans" and "Covered Models". Prices are US dollars per million tokens before tax. Percentage savings are Anthropic's estimates. This briefing is general information, not data-protection advice; confirm your own retention position with Anthropic or your cloud provider.