Intercom is taking an uncommon gamble for a legacy software program firm: constructing its personal AI mannequin.
The 15-year-old huge customer support platform introduced Fin Apex 1.0 on Thursday, a small, purpose-built AI mannequin that the corporate claims outperforms main frontier fashions from OpenAI and Anthropic on the metrics that matter most for buyer help.
The mannequin powers Intercom's current Fin AI agent, which already handles over two million buyer conversations weekly.
In accordance with benchmarks shared with VentureBeat, Fin Apex 1.0 achieves a 73.1% decision charge—the share of buyer points totally resolved with out human intervention—in comparison with 71.1% for each GPT-5.4 and Claude Opus 4.5, and 69.6% for Claude Sonnet 4.6. That roughly 2 share level margin might sound modest, however it's wider than the standard hole between successive generations of frontier fashions.
"In case you're operating massive service operations at scale and also you've received 10 million clients or a billion {dollars} in income, a delta of two% or 3% is a very great amount of shoppers and interactions and income," Intercom CEO Eoghan McCabe advised VentureBeat in a video name interview earlier this week.
The mannequin additionally exhibits vital enhancements in pace and accuracy. Fin Apex delivers responses in 3.7 seconds—0.6 seconds quicker than the next-fastest competitor—and demonstrates a 65% discount in hallucinations in comparison with Claude Sonnet 4.6.
Maybe most hanging for enterprise patrons: it runs at roughly one-fifth the price of utilizing frontier fashions instantly, and is included in Intercom's current "per-outcome"-based pricing construction for its current buyer plans.
What's the bottom mannequin? Does it even matter?
However there's a catch. When requested to specify which base mannequin Apex was constructed on—and its parameter measurement—Intercom declined.
"We're not sharing the bottom mannequin we used for Apex 1.0—for aggressive causes and in addition as a result of we plan to modify base fashions over time," an organization spokesperson advised VentureBeat. The corporate would solely affirm that the mannequin is "within the measurement of lots of of billions of parameters."
For comparability, Meta's Llama 3.1 ranges from 8 billion to 405 billion parameters, and bigger frontier fashions like GPT-5.4 are suspected to be within the trillions.
Whether or not Apex's efficiency claims maintain up towards that context—or whether or not the benchmarks replicate optimizations attainable solely in slim, domain-specific functions—stays an open query.
Intercom says it realized from the backlash AI coding startup Cursor confronted when critics accused the coding assistant of burying the truth that its Composer 2 mannequin was constructed on fine-tuned open-weights fashions relatively than proprietary know-how. However the lesson Intercom drew might not fulfill skeptics: the corporate is clear that it used an open-weights base, simply not which one.
"We’re very clear that now we have" used an open-weights mannequin, the spokesperson stated. But declining to call the mannequin whereas claiming transparency is a contradiction that can probably draw scrutiny—significantly as extra firms tout "proprietary" AI that quantities to post-trained open-source foundations.
Publish-training as the brand new frontier
Intercom's argument is that the bottom mannequin merely doesn't matter a lot anymore.
"Pre-training is sort of a commodity now," McCabe stated. "The frontier, if you’ll, is definitely in post-training. Publish-training is the exhausting half. You want proprietary information. You want proprietary sources of reality."
The corporate post-trained its chosen basis utilizing years of proprietary customer support information gathered by means of Fin, which now resolves 2 million buyer queries per week. That course of concerned extra than simply feeding transcripts right into a mannequin. Intercom constructed reinforcement studying techniques grounded in actual decision outcomes, instructing the mannequin what profitable customer support really appears like—the suitable tone, judgment calls, conversational construction, and critically, methods to acknowledge when a difficulty is actually resolved versus when a buyer continues to be pissed off.
"The generic fashions are skilled on generic information on the web. The precise fashions are skilled on hyper-specific area information," McCabe defined. "It stands to cause due to this fact that the intelligence of the generic fashions is generic, and the intelligence of the precise fashions is domain-specific and due to this fact operates in a far superior means for that use case."
If McCabe is true that the magic is solely in post-training, the reluctance to call the bottom turns into more durable to justify. If the muse is actually interchangeable, what aggressive benefit does secrecy shield?
A $100 million guess paying off
The announcement comes as Intercom's AI-first pivot seems to be working. Fin is approaching $100 million in annual recurring income and rising at 3.5x, making it the fastest-growing section of the corporate's $400 million ARR enterprise. Fin is projected to signify half of Intercom's whole income early subsequent yr.
That trajectory represents a exceptional turnaround. When Fin launched, its decision charge was simply 23%. At present it averages 67% throughout clients, with some massive enterprise deployments seeing charges as excessive as 75%.
To make this occur, Intercom grew its AI staff from roughly 6 researchers to 60 over the previous three years—a big funding for an organization that McCabe admits was "in a very dangerous place" earlier than its AI pivot. The common progress charge for public software program firms sits round 11%; Intercom expects to hit 37% progress this yr.
"We're by far the primary within the class to coach our personal mannequin," McCabe stated. "There's nobody else that's going to have this for a yr or extra."
The speciation and specialization of AI
McCabe's thesis aligns with a broader development that Andrej Karpathy, former AI chief at Tesla and OpenAI, lately described because the "speciation" of AI fashions—a proliferation of specialised techniques optimized for slim duties relatively than common intelligence.
Customer support, McCabe argues, is uniquely fitted to this strategy. It's considered one of solely two or three enterprise AI use circumstances which have discovered real financial traction to date, alongside coding assistants and probably authorized AI. That's attracted over a billion {dollars} in enterprise funding to rivals like Decagon and Sierra—and made the area, in McCabe's phrases, "ruthlessly aggressive."
The query is whether or not domain-specific fashions signify a sturdy benefit or a short lived arbitrage that frontier labs will finally shut. McCabe believes the labs face structural limitations.
"Perhaps the long run is that Anthropic has a giant providing of many various specialised fashions. Perhaps that's what it appears like," he stated. "However the actuality is that I don't suppose the generic fashions are going to have the ability to sustain with the domain-specific fashions proper now."
Past effectivity to expertise
Early enterprise AI adoption targeted closely on price discount—changing costly human brokers with cheaper automated ones. However McCabe sees the dialog shifting towards expertise high quality.
"Initially it was like, 'Holy shit, we are able to really do that for a lot cheaper.' And now they're considering, 'Wait, no, we may give clients a much better expertise,'" he stated.
The imaginative and prescient extends past easy question decision. McCabe imagines AI brokers that perform as consultants—a shoe retailer's bot that doesn't simply reply delivery questions however presents styling recommendation and exhibits clients how completely different choices would possibly look on them.
"Customer support has all the time been fairly shit," McCabe stated bluntly. "Even the perfect manufacturers, you're left ready on a name, you're bounced round completely different departments. There's a possibility now to supply actually excellent buyer expertise."
Pricing and availability
For current Fin clients, the improve to Apex comes at no extra price. Intercom confirmed that buyer pricing stays unchanged—customers proceed to pay per end result as earlier than, at $0.99 per resolved interplay, and robotically profit from the brand new mannequin.
Apex is just not obtainable as a standalone mannequin or by means of an exterior API. It’s accessible solely by means of Fin, that means companies can not license the mannequin independently or combine it into their very own merchandise. That constraint might restrict Intercom's potential to monetize the mannequin past its current buyer base—however it additionally retains the know-how proprietary in a sensible sense, no matter what the underlying base mannequin seems to be.
What's subsequent
Intercom plans to increase Fin past customer support into gross sales and advertising and marketing—positioning it as a direct competitor to Salesforce's Agentforce imaginative and prescient, which goals to supply AI brokers throughout the shopper lifecycle.
For the broader SaaS trade, Intercom's transfer raises uncomfortable questions. If a 15-year-old customer support firm can construct a mannequin that outperforms OpenAI and Anthropic in its area, what does that imply for distributors nonetheless counting on generic API calls? And if "post-training is the brand new frontier," as McCabe insists, will firms claiming breakthroughs face stress to indicate their work—or proceed hiding behind aggressive secrecy whereas touting transparency?
McCabe's reply to the primary query, specified by a current LinkedIn publish, is stark: "In case you can't grow to be an agent firm, your CRUD app enterprise has a diminishing future."
The reply to the second stays to be seen.
