They do have a huge RL budget, but their OG primary source were all books they scanned. Books were a lot cheaper and higher quality than other data. This is straight from their court case.
Everyone else used the same books.
OpenAI: https://news.bloomberglaw.com/ip-law/openai-risks-billions-a...
Google: https://www.theguardian.com/books/2026/jul/14/publishers-sue...
There is zero evidence that book data significantly increased coding performance. It increases coverage of knowledge, but not much else.