October 26, 2026 · 7 min read
Haiku, Sonnet, Opus or Fable: Which Claude Model Fits Which Brokerage Task
Anthropic names four current Claude models and says what each is for. Here is that guidance turned into a decision guide for a brokerage: sorting, drafting, planning and deep analysis, task by task, plus a test to run before anyone upgrades.
Use the smallest Claude model that passes your check. Haiku fits high-volume sorting and extraction, Sonnet fits everyday drafting, Opus fits planning and long, multi-part work, and Fable is for demanding reasoning and long-horizon agentic work. Anthropic's own pages say what each is for. They do not mention real estate, so the task matching below is our reading.
Most brokerages choose the other way. Someone opens the biggest model and uses it for everything. That is a habit, not a decision.
This post is part of our guide to AI for real estate marketing. It covers one question: which of the four current Claude models belongs on which job. We use only what Anthropic's pages say about the models. Where we add a view about real estate work, we say so.
What Anthropic says each model is for
Anthropic's models overview gives each current model one line. When we checked on October 23, 2026, they were Claude Haiku 5.5, Sonnet 5.5, Opus 5.5 and Fable 5.1. Lineups change, so read that page before you set a policy.
Haiku 5.5. "For high-volume, latency-sensitive tasks such as classification, extraction, and routing." Its product page adds chat, voice agents and live support.
Sonnet 5.5. "The best combination of speed and intelligence." Its product page points to well-scoped everyday tasks, including investigation, review and drafting.
Opus 5.5. "For long-running agentic coding and knowledge work." Its product page recommends it as the daily driver for coding and knowledge work.
Fable 5.1. "For demanding reasoning and long-horizon agentic work." Anthropic's choosing a model guide calls it the company's most capable model open to all customers.
"Agentic" means the model works through a multi-step job on its own, using tools. Most brokerage AI work is not that. It is a person asking for a draft and reading the answer.
The decision guide: five questions for any task
Run a task through these in order. Stop at the first model that passes.
Is the output a label, a field or a routing decision? Sorting an agent request into a category, pulling price and open house date out of a listing sheet. Start with Haiku.
Is the output a draft a person will read and edit? A caption set, a listing description from supplied facts, a newsletter. Start with Sonnet.
Does the task need judgment across many inputs, or several linked steps? A quarterly content plan from a year of notes, a recruiting message series that must stay consistent. Start with Opus.
Did Opus, at a higher effort setting, still miss on hard reasoning or long, multi-stage work? Move that one task to Fable.
Will a person act on this without reading it? Then no model is the right answer yet. Add a human check first.
Anthropic describes two ways to start. Efficiency first: begin with Haiku and upgrade only for specific gaps. Capability first: begin with Opus, then move down as the workflow settles. We use both. Sorting and drafting begin on the efficient side. Strategy begins on Opus. Anthropic's overview says that if you are unsure, start with Opus 5.5 for most workloads, which suits one-off work. The five questions are for work you repeat.
Task by task: which model for which brokerage job
Haiku: the sorting and extracting desk
Haiku suits short answers at high volume, such as forty agents sending requests through one form.
Sort each request into listing launch, print, email, social or other.
Pull the fields a launch needs (address, price, agent, photo link, open house date) into a checklist.
Sonnet: the everyday drafting bench
Most brokerage drafting should live here. The work is well-scoped: here are the facts, here is the voice, produce a first version.
A caption set for one listing from a fact sheet you supply.
A listing description built from facts only, then checked. Our post on AI listing descriptions and the fair housing check shows the method.
The monthly newsletter draft.
Opus: the planning desk
Opus fits work with many moving parts.
A quarterly content plan that reconciles listings, recruiting posts and the newsletter calendar.
A review of a month of an office's posts against its voice guide, listing every mismatch.
For a brokerage the phrase to remember is knowledge work: read a lot, decide what matters, produce a plan.
Fable: the exception
Anthropic's choosing guide places Fable in a row of its own: agent sessions that run for hours, multistep deep research, and analysis carried through to a finished document, spreadsheet or deck. It says to move to Fable when evaluations at the xhigh or max effort setting still fall short on that kind of work. Most brokerages will use it rarely, for a deep one-time job such as a positioning study that ends in a leadership deck. It is not the model for captions.
A worked example: one listing launch, three models
Take an invented 14-agent boutique brokerage. An agent submits a listing through the office request form.
Haiku reads the form text, sorts it as a listing launch, fills the checklist fields and flags that the open house date is missing.
Sonnet takes the completed fact sheet and the office voice guide and drafts the Instagram carousel text, the Facebook caption and the property email.
A person checks every claim against the listing page, removes anything about the neighborhood that was not supplied, and approves.
Opus is not in the launch. It shows up on the first Monday of the month, reads the previous month's launches and drafts next month's content plan for the broker to edit.
Anthropic names this shape: an orchestrator that delegates bulk work to lower-cost workers, so most tokens are billed at the lower rate. In a brokerage the orchestrator is a person with a checklist. Put the cheap model on repeated work and the stronger model on monthly judgment.
By the numbers
Anthropic's models overview lists API prices per million tokens, input then output. Haiku 5.5 is from $0.10 and $0.50. Sonnet 5.5 is $2 and $10. Opus 5.5 is $4 and $20. Fable 5.1 is $10 and $50. These show how the models rank on cost. For what your own plan includes, check Anthropic's pricing page.
Metricool's 2026 State of AI in Social Media Study, based on more than 700 social media managers, marketers, creators, freelancers, agency professionals and business owners, found that 95% of social media professionals use AI in their work, including 75% who use it daily. It also found that 37% use more than four AI tools, compared with 14% in 2025.
That last figure is the argument for a short, written model policy. The Metricool respondents are social media professionals, not real estate agents, so read the numbers as a signal about the profession, not a brokerage benchmark.
The test to run before anyone upgrades
Anthropic's choosing guide says "Tuning effort is often a better lever than switching models," and that a good evaluation set is "the most important step in the process." For a brokerage, an evaluation set is a folder.
Pick ten real tasks from last month: five caption sets, two newsletters, two request sorts, one content plan.
Write the check first. For a caption: every fact appears in the fact sheet, the voice matches the guide, the disclosure line is present, nothing describes who should or should not live in the home.
Run all ten on the efficient model. Score each pass or fail.
Move only the failures up one model, or raise the effort setting where one exists, and score again.
Write one line per task type. "Request sorting: Haiku. Caption sets: Sonnet. Quarterly plan: Opus."
An hour of this beats a month of arguing about which model is better.
How Urban Marketing Edge handles this
We use AI to move faster on drafts and variations, and we keep strategy and taste human. The model does a first pass of a caption set, an email or a blog draft, and a person on our team checks it against the client's facts and voice before it reaches the post approval system. Nothing is published unseen.
You can start with the free prompts on the tools page, including the listing description prompt. Plans are month to month, billed in advance, no setup fees.
Questions brokers ask
Which Claude model is best for real estate?
None is best for all of it. Anthropic describes Haiku for high-volume classification, extraction and routing, Sonnet as the best combination of speed and intelligence, Opus for long-running agentic coding and knowledge work, and Fable for demanding reasoning and long-horizon agentic work. Match the model to the task, then test it against your own check.
Should every agent use the biggest model?
No. Anthropic says tuning effort is often a better lever than switching models, and that starting with Haiku is often sufficient for many common applications. A caption set from a fact sheet rarely needs the top tier. Save the strongest model for planning and analysis.
Is Fable worth it for a brokerage?
Rarely, and only for one-off deep work. Anthropic describes Fable 5.1 as for demanding reasoning and long-horizon agentic work. Captions do not fit.
How do I know a cheaper model is good enough?
Test it. Pick ten real tasks, write the pass or fail check first, and run them on the efficient model. Move up only the tasks that fail.
Do these model names change?
Yes. Anthropic's page lists current and legacy models, and the lineup moves. Keep the policy by task type and look up the current model for each type every quarter.
Book a 30-minute strategy call
If your office uses AI in pieces, with no written policy and no check, bring what you have. In 30 minutes we map the simplest plan for your content, themes and monthly cadence, and mark which drafting tasks are worth handing to AI first and which stay with a person. No pitch deck, no pressure. Book a 30-minute strategy call, or see what each plan covers on the plans page first.
