A Filipino legal AI scored 75.8% on Bar questions. GPT-4o scored 24.2%.

Anycase.ai, a Filipino startup that builds legal research software, published a test putting its own system against OpenAI's GPT-4o on 120 questions from the 2024 Philippine Bar Examinations.

Anycase answered 75.8 percent of them correctly. GPT-4o with web search got 24.2 percent. The same model without search got 12.5 percent. The wider gap was in invented sources: GPT-4o with search pointed to cases or rules that do not exist in 32.5 percent of its answers, while Anycase did so in zero percent of its own.

The company says the hard part is not the AI, it is the paperwork. Philippine legal material sits across more than 150 government agencies, in old websites, scanned PDFs and physical files. Anycase says it pays a data runner, a staffer it calls Kuya Ry, to go to government offices and file requests, and that lawyers check each document before it enters the system.

Why it matters

Plenty of Filipinos already ask a chatbot about a labor complaint, a barangay case or a contract they were asked to sign. A general chatbot has no dependable copy of Philippine law to read from, so it fills the gap with text that reads correct. A case number that does not exist, inside a demand letter, is not a small mistake. This test puts a number on how often that happens.

The catch to watch

Anycase ran the test on itself. It names no outside auditor in the paper, and it published the score while selling the product, so treat the figures as the company's own claim until someone independent repeats them. The tool is also built and priced for lawyers, not for a person handling their own case. What the numbers do show clearly is the size of the risk in asking a general chatbot about Philippine law.

The Philippines should not just be a market that imports software built elsewhere. We can build deep, local, domain-specific AI too. Beato Bongco, CEO and co-founder of Anycase.ai