The Question That Started It All

It began with a conversation we kept having internally: why should everyone get the same analytical horsepower for every task? When a portfolio manager is running due diligence on a CHF 50 million acquisition, they need precision. Every decimal matters. But when someone is quickly exploring average rents in a neighborhood before a client call, they need speed and efficiency, not necessarily the computational equivalent of a senior analyst spending three hours on the problem.

The real estate consulting industry has operated on a one-size-fits-all model for decades. You hire a firm, you get their methodology, their pace, their pricing structure. What we are building at Quanthome is different. It is about giving professionals the tools to calibrate their analysis to the task at hand.

Four Models, Four Different Approaches

When we integrated the new model selection feature, we settled on four options, each with its own personality and purpose.

Claude 4.5 Opus is our heavyweight. At 2.5× the standard credit consumption, it is not cheap, but the precision is remarkable. I watched it work through a query last week that perfectly illustrates why structured reasoning matters.

A user asked a seemingly simple question: what is the average rent for apartments within 500 meters of a specific address? For a human analyst, this involves multiple steps. Find the building, get its coordinates, define a geographic zone, query the right database tables, and aggregate the results. What impressed me was watching Opus think through the problem out loud. It first called our building table to get the available fields. Then it filtered to find the specific property and extract its coordinates. When it discovered that dwelling data does not include coordinates directly, it reasoned its way to the solution: query the buildings table within the geographic zone, then access the connected dwelling records through the relationship.

Six tool calls. Multiple database tables. Geographic coordinate math to convert a 500-meter radius into latitude and longitude bounds. And it narrated its reasoning at every step, explaining why it needed to pivot when its first approach hit a dead end. That is not just pattern matching. That is structured problem-solving. For critical analyses where mistakes have real financial consequences, Opus is worth every credit.

Claude 4.5 Sonnet became our recommended default for good reason. At 1.5× consumption, it hits a sweet spot that works for most daily use cases. It is intelligent enough to handle sophisticated queries, efficient enough that you will not burn through your monthly allocation in a week. When I'm testing new features or running standard market analyses, Sonnet is my go-to.

Claude 4.5 Haiku is for speed and volume. At just 0.5× consumption, you can run exploratory analyses all day without thinking twice about credits. First pass on a new market? Haiku. Quick question about property classifications? Haiku. It is not going to write your board presentation, but it will help you figure out what questions to ask.

Gemini 3 Pro rounds out the selection at 1.0× consumption. We believe in model diversity, different architectures sometimes catch things others miss, or approach problems from useful angles. Having an alternative keeps us honest and gives users options.

Building for an Industry in Motion

Here is what I keep thinking about as we ship these updates: the pace of change is unprecedented. Six months ago, the models we're now calling "legacy" were state of the art. Six months from now, we will probably be integrating capabilities that don't exist yet today.

This reality shapes how we architect everything. Our model integration layer is designed for flexibility, adding a new model takes days, not months. Every update gets validated against our specific real estate use cases because what works for general conversation does not always work for structured property analysis. The tools that Opus handles elegantly, our aggregation functions, our geographic filters, our market data integrations, required careful prompt engineering and testing.

The traditional consulting model in real estate is built on relationships, intuition, and periodic deep-dive reports. That model is not disappearing, but it is being augmented by something faster, more continuous, and more accessible. A property manager who previously waited weeks for a market analysis can now run one in minutes. An investor who relied on quarterly reports can now check market conditions before any significant decision.

We are not replacing human expertise. We are amplifying it. And with model selection, we are letting users decide exactly how much amplification they need for each task.

Looking Forward

What excites me most is not any single analysis. It is watching the AI reason through problems it hasn't seen before. That geographic radius query I mentioned earlier? We did not explicitly program that workflow. Opus figured out the coordinate math, realized it needed to traverse table relationships, and adapted when its first approach did not work. That kind of flexible problem-solving is what separates a useful tool from a transformative one.

We are building infrastructure for a future where every real estate professional has access to institutional-grade analytical capabilities. Not just the large funds with teams of analysts, but independent consultants, small property managers, individual investors.

Model selection is one step in that direction. It acknowledges that different situations call for different tools, and puts that choice in the user's hands. As new models emerge, and they will, faster than any of us expect, we will integrate them. As capabilities expand, we will expose them.

The real estate industry has been slower than others to embrace technological change. But once transformation starts, it tends to accelerate. We are building for that acceleration, and honestly, it is the most exciting work I have ever done.

The model selection feature is now live for all Quanthome users. Access it through the Model panel on the right side of your chat interface.