In this guide · 8 sections
- 01What Anthropic released in September and October 2026
- 02API prices for Opus 5.5, Sonnet 5.5 and Haiku 5.5
- 03Which Claude plans include the new models
- 04What published benchmarks say
- 05Haiku or Opus for an FAQ chatbot?
- 06How to run your own 30-minute test
- 07Do it yourself or get help?
- 08Frequently asked questions
Short answer
Opus 5.5 (claude-opus-5-5, 22 Sep 2026) costs US$4 / US$20 per million input/output tokens; Haiku 5.5 (claude-haiku-5-5, 7 Oct 2026) costs US$0.10 / US$0.50 for prompts up to 100k tokens. Use Haiku for FAQ bots and sorting; pay for Opus for long reasoning, documents and agents.
What Anthropic released in September and October 2026
Anthropic's Claude release notes list three launches in the new Claude 5.5 family. The models overview now says to start with Claude Opus 5.5 for most workloads, and to use Claude Fable 5.1 only for the most demanding reasoning and long agentic work.
| Model | Launched | API model ID | Anthropic's description |
|---|---|---|---|
| Claude Opus 5.5 | 22 Sep 2026 | claude-opus-5-5 | Long-running agentic coding and knowledge work |
| Claude Sonnet 5.5 | 28 Sep 2026 | claude-sonnet-5-5 | Best combination of speed and intelligence |
| Claude Haiku 5.5 | 7 Oct 2026 | claude-haiku-5-5 | High-volume, latency-sensitive tasks such as classification, extraction and routing |
Anthropic's launch note claims Opus 5.5 performs at the level of Fable 5.1 on most work and costs 40% less to run than Opus 5. All three models have a 1 million token context window, up to 128,000 output tokens and a reliable knowledge cutoff of June 2026, according to the models overview. Anthropic lists retirement as not sooner than September 2027 for Opus 5.5 and October 2027 for Haiku 5.5.
API prices for Opus 5.5, Sonnet 5.5 and Haiku 5.5
These prices come from Anthropic's API pricing page, in US dollars per million tokens (MTok), as of October 2026. A token is roughly three-quarters of an English word.
| Model | Input | Output | Cache hit (input) | Batch API (input / output) |
|---|---|---|---|---|
| Opus 5.5 | US$4 | US$20 | US$0.20 | 50% off |
| Sonnet 5.5 | US$2 | US$10 | US$0.10 | US$1 / US$5 |
| Haiku 5.5, prompt up to 100k tokens | US$0.10 | US$0.50 | US$0.01 | US$0.05 / US$0.25 |
| Haiku 5.5, prompt over 100k tokens | US$0.50 | US$2.50 | US$0.05 | US$0.25 / US$1.25 |
| Fable 5.1 (for comparison) | US$10 | US$50 | US$0.25 | 50% off |
Two details matter for budgeting. Haiku 5.5 is the only current model priced by prompt length: a single request over 100,000 tokens pays five times the rate. And Opus 5.5 also has a fast mode at US$8 / US$40, which you rarely need for business chat. For a wider comparison with OpenAI and Google, see our LLM API pricing per token guide.
Which Claude plans include the new models
If your staff use the Claude app rather than the API, the Claude pricing page shows which model families each plan includes. The newest version in each family is what users see.
| Plan (Oct 2026) | Price | Opus | Sonnet and Haiku |
|---|---|---|---|
| Free | US$0 | No | Yes |
| Pro | US$17/month annual, US$20 monthly | Yes | Yes |
| Max | From US$100/month | Yes | Yes |
| Team, Standard seat | US$20/seat/month annual, US$25 monthly | Yes | Yes |
| Team, Premium seat | US$100/seat/month annual, US$125 monthly | Yes | Yes |
Anthropic's 7 October release note also says Max and Team plans now include monthly API credits for building your own apps, claimed by linking a Claude Console organisation; the pricing page shows up to US$500 a month in pooled credits on Team. Our Claude Team plan guide covers seat types, admin controls and data handling in detail.
What published benchmarks say
We have not run our own tests. The figures below are from Artificial Analysis, an independent benchmarking site, as retrieved on 20 October 2026. Its Intelligence Index combines several reasoning, knowledge and coding tests; cost per task reflects how many tokens a model uses at a given effort setting.
| Model and effort | Intelligence Index | Cost per task | Output speed |
|---|---|---|---|
| Opus 5.5, medium (default) | 51 | US$1.34 | 81 tokens/s |
| Opus 5.5, high | 54 | US$1.82 | 82 tokens/s |
| Haiku 5.5, medium (default) | 34 | US$0.05 | 179 tokens/s |
Read it simply: Opus is clearly smarter on hard tasks, and Haiku is around 25 times cheaper per task and about twice as fast. Most small business tasks, such as answering a question from your FAQ, sorting an email or pulling a date out of an invoice, do not need the top score.
Haiku or Opus for an FAQ chatbot?
A website assistant that answers questions from your own price list, opening hours and policies mostly needs to find the right paragraph and say it clearly. That is the kind of high-volume, retrieval-style work Anthropic designed Haiku for.
Haiku 5.5 is usually enough when
- Answers come from your FAQ, price list or policies
- Many short conversations a day
- Fast replies matter more than deep reasoning
- Sorting enquiries, tagging leads, extracting fields
Consider Opus 5.5 when
- Answers need judgement across long documents or contracts
- Multi-step tasks such as drafting a proposal from notes
- Agent work that calls several tools in sequence
- Mistakes are costly, for example quoting complex prices
A practical approach is to start on Haiku, log the questions it gets wrong, and only route those question types to Sonnet or Opus. Our guide to Claude vs ChatGPT for business writing covers the writing side if staff mainly use Claude to draft emails and documents.
How to run your own 30-minute test
Benchmarks are a guide, not a decision. A short test on your own material tells you more.
Collect 20 real questions
Take them from your enquiry inbox or chat history, including a few awkward ones. Remove names and contact details first.
Write one system prompt
Paste in your FAQ, prices and policies, and tell the model to say "I don't know" when the answer is not in the text.
Run the same questions on Haiku 5.5 and Opus 5.5
Use the Claude Console workbench or a simple script, with the same prompt and settings.
Score each answer: right, partly right or wrong
Have someone who knows the business mark them. Note any invented prices or policies.
Compare cost and speed
Check token usage in the Console. If Haiku gets nearly all of them right, use it and keep Opus for the exceptions.
Do it yourself or get help?
Switching staff to the new models in the Claude app needs no work: they appear automatically on plans that include them. Building a chatbot or automation on the API is where model choice, prompts, logging and cost limits matter.
Our AI Website Assistant (S$790) builds a chatbot that answers from your own FAQ, price list and policies, on a model sized for the job. For back-office tasks such as sorting enquiries or extracting invoice data, Workflow Automation (S$390 per workflow) connects an AI step to your existing tools. IT Sorted is a Singapore-based team of developers with over 10 years of experience. You pay upfront, send your brief in writing through the site chat or contact form, and track progress on your order page.
Frequently asked questions
Anthropic's release notes list Claude Opus 5.5 on 22 September 2026, Claude Sonnet 5.5 on 28 September 2026 and Claude Haiku 5.5 on 7 October 2026. All three are available on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry, according to Anthropic's models overview, and in the Claude apps on plans that include each model family.
Anthropic's pricing page lists Opus 5.5 at US$4 per million input tokens and US$20 per million output tokens as of October 2026, down from US$5 and US$25 for Opus 5. Cache hits cost US$0.20 per million tokens and the Batch API gives 50% off. Fast mode costs US$8 and US$40 per million tokens.
For most FAQ-style chatbots, yes. Anthropic positions Haiku 5.5 for high-volume, latency-sensitive tasks, and it costs US$0.10 per million input tokens for prompts up to 100,000 tokens. It scores lower than Opus on reasoning benchmarks, so test it on 20 real customer questions and route only the hard question types to a larger model.
No. Claude's pricing page, checked in October 2026, shows the Free plan includes Sonnet and Haiku models but not Opus. Opus is included on Pro, Max, Team and Enterprise plans. Team Standard seats cost US$20 per seat per month billed annually, and Pro costs US$17 a month billed annually.
Not always. Anthropic's API release notes say Opus 5.5 always uses adaptive thinking and returns errors if code tries to disable thinking or force a specific tool. Code written for Haiku 4.5 can break on Haiku 5.5 if it sets temperature, top_p or top_k, or uses assistant message prefill. Test in a staging environment before changing the model ID.