Which AI Models Are Actually Being Used in Commercial Real Estate Right Now
A Q1 2026 snapshot of foundation model adoption across CRE workflows -- what teams are choosing and why.
Updated March 22, 2026. This landscape is evolving rapidly. We refresh this post quarterly. Next update: Q2 2026.
The foundation model conversation in commercial real estate has moved past theory. Institutional development teams, REIT development groups, and CRE analytics firms are actively selecting and deploying specific models for specific workflows. The choices are not arbitrary -- they reflect real differences in performance, cost, context window, and specialization.
Here is a practical survey of what is in active use and why.
The Three Providers Dominating CRE Enterprise Deployments
Three providers account for the majority of foundation model usage in institutional CRE: Anthropic (Claude), OpenAI (GPT-5.4), and Google (Gemini). Each has a distinct performance profile.
Anthropic Claude Opus 4.6
Claude has extended its lead across the most demanding CRE workflows. Released in February 2026, Claude Opus 4.6 is Anthropic's strongest model to date and has been publicly benchmarked as industry-leading across tool use, search, and finance -- the last of which maps directly to CRE underwriting, pro forma modeling, and IRR analysis tasks.
Its 200,000-token context window continues to allow full commercial leases, loan agreements, and environmental reports to be processed in a single pass. Development teams using Claude for lease abstraction, PSA review, and Phase I summarization consistently report fewer hallucinations on structured legal content. Anthropic's training approach produces cautious, qualified outputs -- useful when a document is ambiguous and you need the model to flag uncertainty rather than fabricate certainty.
Claude Opus 4.6 is now the dominant choice for both document review and financial modeling at firms investing in AI-assisted workflows.
OpenAI GPT-5.4
GPT-5.4 is OpenAI's current flagship, released in March 2026. It replaces GPT-4o as the standard for professional and enterprise deployments and brings meaningful gains in structured reasoning and multi-step inference.
It remains the most widely integrated model across third-party CRE platforms, largely because of OpenAI's head start on enterprise API relationships. GPT-5.4 excels at structured data extraction -- rent rolls, market comparables, financial statement parsing -- and is the default in several established CRE analytics platforms that have not yet updated their model routing.
The reasoning capability improvement over GPT-4o is material for complex underwriting tasks, though teams evaluating head-to-head against Claude Opus 4.6 on finance-specific tasks are increasingly finding Claude competitive or ahead.
Google Gemini 1.5 Pro / 2.0
Gemini 1.5 Pro's 1 million-token context window makes it the only model capable of handling a full data room in a single inference call. For development teams reviewing dozens of documents simultaneously -- or for pipeline monitoring workflows that need to process weeks of emails and reports together -- Gemini's scale advantage is real.
Adoption is growing fastest in teams with existing Google Cloud infrastructure. The integration path is cleaner and the enterprise support relationship is already in place.
Adoption by Workflow Category
Document review and extraction: Claude Opus 4.6 leads. Context window, accuracy on legal text, and hallucination rate on structured documents are the deciding factors.
Financial modeling and analysis: Claude Opus 4.6 is now competitive with or ahead of GPT-5.4 for pro forma automation, IRR sensitivity analysis, and multi-step underwriting tasks. Teams that have updated their model routing in early 2026 are predominantly running Claude here. GPT-5.4 remains common on platforms that haven't yet updated their defaults.
Market research synthesis: Gemini 1.5 Pro for large-volume document ingestion. Claude Opus 4.6 for higher-quality synthesis when the document set is defined and bounded.
Development pipeline reporting: Mixed. Most workflow platforms abstract the model layer, so teams use whatever their platform defaults to. Claude and GPT-5.4 are both common.
Construction monitoring (image analysis): GPT-5.4 for site photo analysis and schedule comparison. Gemini for cross-referencing large image archives against as-built documentation.
Site screening and GIS analysis: GPT-5.4 via function-calling APIs for structured data retrieval. Claude for synthesizing the output into decision-ready briefs.
How Teams Are Making Model Decisions
Three criteria consistently drive model selection at institutional development firms.
Task-specific accuracy, not benchmark scores. Generic benchmarks (MMLU, HumanEval, GPQA) do not predict performance on CRE documents. Firms that invest in task-specific evaluations -- running a sample of their own documents through candidate models and comparing outputs -- make significantly better decisions than those relying on published scores.
Cost per workflow. Enterprise pricing varies by provider and volume tier. A firm running thousands of lease abstractions monthly has a materially different cost profile with Claude Opus 4.6 versus GPT-5.4 versus a distilled open-source model. The arithmetic changes the right answer.
Integration depth. The model that connects most cleanly to your existing data stack often delivers more value than the marginally better model that requires significant engineering. API reliability, uptime SLAs, and enterprise support tier matter operationally.
What's Still Early
Multi-agent coordination. Multiple AI models working on different workflow steps and passing structured results between them is being tested but is not production-ready at most firms. Coordination overhead and error propagation make it more fragile than a capable single model with well-structured retrieval.
CRE-specific fine-tuned models. Several platforms are developing fine-tuned models on CRE-specific document corpora. Early results suggest meaningful accuracy gains for specific document types -- lease abstraction, Phase I summarization, rent roll extraction. Expect this to become a differentiated capability by Q3 2026.
On-premises and private cloud deployment. Regulatory pressure and data privacy concerns are driving interest in private deployments of open-source models (Llama 3, Mistral). Most institutional deployments still rely on API-based access to frontier models. Private deployment at scale remains a 12-18 month horizon for most firms.
The Honest Assessment
No single model is best for all CRE use cases. The firms extracting the most value are not locked into one provider. They are using different models for different workflow steps, often abstracted behind a platform layer that allows them to swap models as capabilities and pricing evolve.
The model landscape shifted materially in early 2026. Claude Opus 4.6 raised the ceiling on what is achievable for finance and document-heavy tasks. GPT-5.4 brings OpenAI's flagship into the current generation. Teams that locked in model choices in 2025 and haven't re-evaluated are likely leaving accuracy and efficiency on the table.
The decision is less 'which model' and more 'which model for what task, at what cost, with what integration, and with what fallback when it is wrong.'
Teams that build workflows with human review checkpoints at high-stakes decision points -- rather than treating AI output as terminal -- are deploying faster and with fewer costly errors.
Model capabilities and pricing are advancing faster than publication cycles. Check with your platform providers for current benchmarks and pricing. This post reflects publicly available information and practitioner observations as of March 2026.