What Is the Best LLM? It Depends. And Your Agents May Be Paying for Too Many.
Everyone asks, “What’s the best LLM?” They probably ask what is the best AI. I think that most folks know that AI stands for artificial intelligence, but AI really refers to the term LLM or large language model.
This is usually followed by, “We should probably standardize on one,” and then, quietly, “Do we know what this costs?”
The short answer is simple.
There is no single best LLM.
There is only the best LLM for the job, for the moment, and for who is paying the bill.
And right now, in many brokerages, that bill is quietly landing on agents.
There are three popular LLMs or AI that are popular with agents and each has its own superpower: ChatGPT the writer, Gemini the image designer, and Claude the programmer.
ChatGPT: The One That Writes Like You and Remembers the Entire Project
ChatGPT has a real advantage in writing, but not for the reason most people think.
It’s not just fluent. It’s persistent.
With Projects, ChatGPT stops behaving like a chatbot and starts behaving like a workspace. Strategy, drafts, revisions, objections, and decisions all live together. The work compounds instead of resetting.
“Projects are where ChatGPT stops being a tool and starts behaving like a teammate who actually remembers the last meeting.”
— Victor Lund, WAV Group
Why this matters to brokers and MLS CEOs:
Strategy is not a single document. It’s a sequence. Projects allow leadership teams to develop positioning, test language, refine messaging, and keep everything aligned over time.
Example:
An MLS executive creates a ChatGPT Project around a market-wide policy rollout. Inside are background documents, broker FAQs, draft emails, talking points, and revisions. Over weeks, ChatGPT helps refine explanations, anticipate pushback, and keep the tone consistent.
At some point, it stops feeling like prompting and starts feeling like collaboration.
“It’s the difference between asking random questions and running an organized campaign.”
— Victor Lund
Small warning label:
Projects are powerful, but running every task on the most advanced model is like bringing a senior partner into a meeting to take notes. Effective AI strategies match the model to the task, not the ego.
Gemini Banana: The One That Thinks in Pictures and Skips the Instructions
Gemini Banana is not the model for precision.
If you ask for exact changes, it may respond with something more philosophical than architectural. That’s not a flaw. That’s the point.
Gemini Banana excels at visual exploration, especially when you start with real photos.
Example:
An agent photographs a listing and asks Gemini to imagine it with a fenced in yard for the buyer’s dog. The output is not real, but it is perfect for starting a conversation with buyer to imagine how this home with the fence will look.
“Gemini Banana is the fastest way I know to turn a vague idea into something someone else can react to.”
— Kevin Hawkins
This is brainstorming, not engineering. And brainstorming does not need a premium model every time unless your inspiration budget is unlimited. Don’t forget – in California beginning in January, using AI to modify an image used in real estate marketing requires that you watermark the image to convey that it was altered and link or include the original photo.
Claude: The Serious One Who Actually Reads the Assignment
Claude is pulling ahead in coding and technical reasoning.
It reads long instructions. It thinks through edge cases. It behaves like the person on the team who quietly prevents disasters.
Example:
A real estate technology team uses Claude to design workflows connecting MLS data, CRMs, and agent dashboards. Claude flags risky assumptions, identifies validation points, and produces logic that works in production.
“Claude is the model you use when the output feeds another system, not just a document.”
— David Gumpper
Claude is not flashy. It is dependable. That distinction matters when mistakes show up in live environments.
The Quiet Problem: Agents Are Buying Too Many AIs
Here’s the part most brokerages and MLSs are not talking about yet.
Agents are licensing multiple LLMs on their own.
ChatGPT. Claude. Gemini. Sometimes more.
At today’s rates, an agent subscribing to three LLM platforms can easily spend $60 or more per month, personally, just to stay competitive. Multiply that across a brokerage, and the inefficiency becomes obvious.
“If your agents are each paying for three AIs, you don’t have an AI strategy. You have real estate agents looking outside for AI solutions that you could be fulfilling.”
— Victor Lund
The Brokerage Opportunity: AI as a Shared Advantage
Forward-thinking brokerages are already solving this.
At firms like Seven Gables, agents access AI solutions through the brokerage, not through individual subscriptions. The brokerage provides AI agents trained for specific needs, such as:
- Professional biographies
- Personal branding content
- AEO (Answer Engine Optimization) analysis
- Market-specific messaging
Agents get expert-level AI without managing multiple subscriptions or paying out of pocket. The brokerage gains consistency, governance, and leverage.
This is not about saving agents money, although it does.
It is about positioning the brokerage as the platform where intelligence lives.
Tokens, Versions, and the Bill That Eventually Shows Up
Model choice matters.
Model version matters more.
Token cost matters most at scale.
As we outlined in a recent article, AI Token Costs Are Invisible Until They Aren’t, premium models can cost 10 to 30 times more than smaller versions when deployed broadly.
Using the most advanced model for every task is like hiring an executive to answer the phone. Impressive, but unnecessary.
The Strategy (And the Punchline)
The best LLM strategy is not picking a favorite model.
It is knowing:
- When ChatGPT Projects make sense
- When Claude should do the reasoning
- When Gemini Banana should do the imagining
- And when a smaller, cheaper model is more than enough
“The winners in AI won’t be the ones with the smartest model. They’ll be the ones who knew when not to use it.”
— Marilyn Wilson
For brokers and MLS CEOs, AI is no longer a novelty. It is infrastructure. And like all infrastructure, it works best when it is shared, intentional, and designed for the people who rely on it every day.
“Agents are solving real problems. They’re just doing it alone. At Seven Gables, agents access AI through the brokerage instead of paying for multiple tools themselves.”
— Victor Lund
The Executive Gut Check
If your agents are each paying $60/month for AI…
- That’s not innovation
- That’s fragmentation
“This is already happening. The only question is whether agents keep paying for AI individually, or whether the brokerage becomes the place where intelligence lives.”
— Victor Lund
