The sales pitch for industry-specific AI has been the same for two years. General chatbots are fine for everyday stuff, but if you want serious work done in medicine, law, or finance, you need a specialized tool trained on your industry's data. The data is starting to say otherwise.
A new study in Nature Medicine tested the top general-purpose models against the clinical AI tools hospitals actually pay for. The result was rough for the specialists. Gemini 3.1 Pro hit 97.4% accuracy on MedQA, the standard medical licensing benchmark. GPT-5.2 hit 94.2%. Both beat OpenEvidence (89.6%) and UpToDate (88.4%), the two clinical decision tools that doctors have been subscribing to for years.
The uglier finding came on real clinical questions, where 12 blinded physicians rated the answers. Frontier models clustered at the top. OpenEvidence and UpToDate landed in a second tier, scoring no better than Google's AI Overview, the free summary that pops up above search results. UpToDate's tool also refused to answer 19% of questions, compared to 1-3% for everything else.
For context, OpenEvidence just raised $210 million at a $3.5 billion valuation, calling itself the fastest-growing physician app in history. It's now benchmarking at rough parity with a feature Google gives away for free.
The same pattern is showing up in other verticals where specialized AI was supposed to be untouchable.
- Legal: Harvey, the legal AI startup that's raised hundreds of millions on the promise of beating general models on lawyer work, just published results from a preview of OpenAI's GPT-5.5. The model showed real gains on legal reasoning and document structure, according to Niko Grupen, Harvey's head of applied research. Harvey isn't training its own frontier model. It's wrapping the best one OpenAI ships.
- Finance: Hebbia, the finance AI company, is now leaning on Claude Opus 4.6 for the document-heavy work its analysts do all day. "Creating financial PowerPoints that used to take hours now takes minutes," CTO Aabhas Sharma told Anthropic in a customer writeup.
The buyer side is catching up to what the benchmarks show. Adar Pallis, who runs clinical applications and technology at Providence, one of the largest hospital systems in the US, said the goal now is to reduce the number of AI apps in their stack, not add more. Fewer vendors, fewer logins, fewer integration headaches. That's a very different conversation than the one specialized AI startups were having two years ago, when every hospital, law firm, and bank felt like they had to buy a vertical tool to keep up.
Gartner's Arunasree Cheparthi has argued that domain-specific models still win on cost and relevance for narrow use cases, and that's probably true at the margins. But the margins keep shrinking every time OpenAI, Google, or Anthropic ships a new model. A recent CIO.com analysis put it bluntly: AI has changed the economics of writing software, which means the moat for a lot of these vertical companies was never the code or the training data. It was the cost of building it in the first place. That cost is collapsing.
The vertical AI boom was built on the assumption that the frontier labs would stay generalist and leave the industry-specific work to specialists. That bet is looking worse every quarter. When a free Google search feature can match a $3.5 billion clinical AI startup on physician-rated answers, the question for anyone selling specialized AI isn't whether they're better than ChatGPT today. It's whether they'll still be better six months from now, when the next general-purpose model ships. Most of them won't be. The vertical AI companies that survive this stretch will be the ones that figured out early they're not actually in the AI business. They're in the workflow, distribution, and trust business, and they happen to use someone else's model to do it.
