HubSpot’s Breeze uses LLMs
Reporting says HubSpot’s Breeze product layers third‑party large language models over CRM workflows, running core agents on GPT‑4.x while migrating marketplace agents to GPT‑5 in early 2026 and applying OpenAI reasoning to customer data. The coverage frames many CRM 'AI' features as integrations of foundation models with proprietary data and workflows. (sourceforge.net)
HubSpot’s Breeze is not a single model built for marketing. It is a layer of large language models wired into HubSpot’s customer-relationship-management workflows and data. (sourceforge.net) SourceForge reported on April 15, 2026, that HubSpot’s core Breeze agents run on GPT-4.x, while Breeze Studio marketplace agents moved to GPT-5 in early 2026. The same report said Breeze uses OpenAI reasoning on customer data rather than a model trained on each customer’s own campaign history. (sourceforge.net) HubSpot’s own product pages describe Breeze as “HubSpot’s AI” across marketing, sales, and service, and Breeze Studio as the place to discover, customize, and deploy agents. HubSpot’s knowledge base says assistants answer questions and generate content, while agents carry out structured tasks with account data and tools. (hubspot.com 1) (hubspot.com 2) (knowledge.hubspot.com) That setup reflects how much business software now uses artificial intelligence: the model writes or reasons, and the software company supplies the customer records, permissions, and workflow rules. HubSpot said in a June 26, 2025 blog post that Breeze is built in layers, with unified structured data like customer-relationship-management records and unstructured data like emails, calls, and transcripts at the base. (blog.hubspot.com) HubSpot has pushed that architecture further in 2026 with a Model Context Protocol server that lets compatible assistants query and write HubSpot customer-relationship-management data through natural-language requests. HubSpot’s developer docs say that server gives artificial-intelligence tools secure, granular access to contacts, deals, engagements, and other records through the company’s application programming interfaces. (developers.hubspot.com 1) (developers.hubspot.com 2) The company is also tying HubSpot data directly into OpenAI products. HubSpot announced this month that it launched a deep research connector for ChatGPT, and a separate HubSpot page says the connector is available to HubSpot customers on all tiers if they also have an OpenAI account. (hubspot.com 1) (hubspot.com 2) HubSpot’s trust materials say it uses both proprietary models and third-party providers for generative artificial intelligence features. Its cloud infrastructure frequently asked questions say HubSpot does not permit those providers to use customer data for model training and seeks zero-day retention where possible. (trust.hubspot.com) (knowledge.hubspot.com) HubSpot is also explicit that access depends on product tier, permissions, and credits. Breeze Studio documentation updated March 17, 2026 says some features require HubSpot credits, and HubSpot’s general Breeze documentation says some capabilities also require specific seats or additional subscriptions. (knowledge.hubspot.com) (knowledge.hubspot.com) The result is that “AI in the customer-relationship-management” often means a foundation model from one company, wrapped in another company’s data model, billing system, and automation stack. Breeze’s pitch is less about inventing a new model than about putting GPT-style systems inside the parts of HubSpot where sales, marketing, and service teams already work. (sourceforge.net) (blog.hubspot.com)
