Intercom is taking an unusual gamble for an old software company: building its own AI model.
A 15-year-old giant customer service platform based in Dublin, Ireland Fin has announced Apex 1.0 on Thursday, the company claims its small, purpose-built AI model outperforms OpenAI and Anthropic’s leading edge models on metrics most important to customer support.
The model gives power Intercom’s existing Finnish AI agentalready handling over one million customer conversations per week.
According to benchmarks shared with VentureBeat, Fin Apex 1.0 achieves a 73.1% resolution rate for the percentage of customer issues that are fully resolved without human intervention, compared to 71.1% for both GPT-5.4 and Claude Opus 4.5, and 69.6% for Claude Sonnet 4. This roughly 2 percentage point margin may sound modest, but it is wider than the typical gap between successive generations of frontier models.
"If you’re running a large service operation at scale and you have 10 million customers or a billion dollars in revenue, a 2% or 3% delta is a really big customer, interaction, and revenue." Intercom CEO Eoghan McCabe told VentureBeat in a video call interview earlier this week.
The model also shows significant improvements in speed and accuracy. The Fin Apex delivers responses in 3.7 seconds – 0.6 seconds faster than the next fastest competitor – and the Claude Sonnet shows a 65% reduction in hallucinations compared to 4.6.
Perhaps most surprising for enterprise buyers: it costs about one-fifth of using direct border models and is included in Intercom’s existing tariffs. "as a result"reasonable pricing structure for existing customer plans.
What is the basic model? Does it even matter?
But there is a catch. When asked to specify which base model the Apex is based on and its parameter size, Intercom declined.
"We don’t share the base model we used for Apex 1.0—for competitive reasons and also because we plan to change base models over time," a company spokesperson told VentureBeat. The company will only confirm that the model is available "hundreds of millions of parameters in size."
This is a particularly small model. By comparison, Meta’s Llama 3.1 ranges from 8 billion to 405 billion parameters; Even effective open-weight models like the Mistral 7B are dwarfed by the billionths of a magnitude Intercom describes.
Whether Apex’s performance claims are appropriate in this context—or whether the benchmarks reflect only the optimizations possible in narrow, domain-specific applications—remains an open question.
He says he learned from Intercom AI coding startup faced Cursor when critics accused the coding assistant of hiding the fact that its Composer 2 model was built on fine-tuned open-weight models, not proprietary technology. But Intercom’s lesson may not satisfy skeptics: the company is transparent about using an open weight base, just not. which one.
"We are very transparent" a spokesperson said it uses an open weight model. Refusing to name the model while claiming transparency is a contradiction that will likely attract scrutiny — especially as more companies chime in. "owner" AI equivalent to open source foundations after training.
Post-training as the new frontier
Intercom’s argument is that the underlying model simply doesn’t matter anymore.
"Advance preparation is now a commodity," McCabe said. "The frontier, if you will, is actually post-training. It’s the hard part after training. You need private information. You need private sources of truth."
The company has trained on its chosen foundation, using years of dedicated customer service data gathered through Fin, which now handles 2 million customer inquiries per week. This process involved more than just feeding transcripts into a model. Intercom built reinforcement learning systems based on real solving results, teaching the model what successful customer service really looks like—appropriate tone, judgment calls, conversational structure, and, critically, how to recognize when a problem is actually solved and when a customer is still frustrated.
"General models are trained on general data on the internet. Specific models are trained on hyper-specific domain data," McCabe explained. "Therefore, it stands to reason that the intelligence of generic models is generic and the intelligence of specific models is domain specific and therefore performs superiorly for this use case."
If McCabe says the magic is entirely after training, it becomes hard to justify not wanting to name the base. If the foundation is truly interchangeable, what competitive advantage does privacy preserve?
A $100 million bet pays off
The announcement comes as Intercom’s first AI-powered pivot goes live. Fin is approaching $100 million in annual recurring revenue, growing 3.5x, making it the fastest growing segment of the company’s $400 million ARR business. Fin is predicted to account for half of Intercom’s total revenue early next year.
This trajectory represents a remarkable turnaround. When Fin was launched, its resolution rate was only 23%. Today, it averages 67% among customers, with some large enterprise deployments reaching 75%.
To make that happen, Intercom has grown its AI team from six researchers to 60 over the past three years — a significant investment for the company, McCabe acknowledged. "in a really bad place" Before the AI loop. The average growth rate for public software companies is about 11%; Intercom expects 37% growth this year.
"We are the first in the category to train our model," McCabe said. "No one else will have it for a year or more."
Specification and specialization of AI
McCabe’s thesis coincides with a broader trend recently described by Tesla and OpenAI’s former AI leader Andrej Karpathy. "specification" the proliferation of artificial intelligence models – the proliferation of specialized systems optimized for narrow tasks rather than general intelligence.
McCabe argues that customer service is uniquely suited to this approach. It’s one of two or three enterprise AI use cases that have found real economic traction so far, along with coding assistants and potentially legal AI. It attracted more than a billion dollars in venture funding to competitors like Decagon and Sierra and, in McCabe’s words, created the space. "fierce competition."
The question is whether domain-specific models represent a lasting advantage or a temporary arbitrage to which frontier labs will eventually close. McCabe believes that labs face structural limitations.
"Perhaps the future is for Anthropic to have a large offering of many different specialized models. Maybe it seems like" he said. "But the reality is that I don’t think generic models can keep up with domain-specific models right now."
Apart from efficiency for practice
Early adoption of AI in enterprises focused on cost reduction—replacing expensive human agents with cheaper automated agents. But McCabe sees the conversation shifting to the quality of experience.
"In the beginning, it was like, “Holy shit, we can do this for less.” Now they’re thinking, ‘Wait, no, we can provide a better customer experience.’" he said.
The vision goes beyond a simple query solution. McCabe envisions AI agents acting as advisors—not just answering delivery questions, but also offering style tips and showing customers how different options might look on them.
"Customer service has always been very bad," McCabe said bluntly. "Even with the best brands, you are left waiting for a call, wandering between different departments. Now you have the opportunity to deliver a truly stellar customer experience."
Price and availability
Upgrading to Apex is free of charge for existing Finnish customers. Intercom confirmed that customer pricing remains unchanged – users continue to pay $0.99 per result per resolved interaction, as before, and automatically benefit from the new model.
Apex is not available as a standalone model or through an external API. It is only available through Fin, meaning businesses cannot license the model independently or integrate it into their own products. This restriction may limit Intercom’s ability to roll out the model beyond its existing customer base, but it also keeps the technology in a practical sense proprietary, regardless of what the underlying model is.
What’s next
Intercom plans to expand Fin beyond customer service into sales and marketing, positioning it as a direct competitor to Salesforce’s vision of Agentforce, which aims to provide AI agents throughout the customer lifecycle.
For the broader SaaS industry, Intercom’s move raises troubling questions. If a 15-year-old customer service company can build a model that outperforms OpenAI and Anthropic in its domain, what does that mean for vendors who still rely on generic API calls? And if "post training is the new frontier," As McCabe insists, will companies that claim to be making progress face pressure to show their work, or will they continue to hide behind competitive secrecy while promoting transparency?
McCabe’s answer to the first question, mentioned in a recent post on LinkedInis sharp: "If you cannot become an agent company, the future of your CRUD application business is diminished."
The answer to the second remains to be seen.
