Hacker News Reader

For context: R1 is a reasoning model based on V3. DeepSeek has claimed that GPU costs to train V3 (given prevailing rents) were about $5M.

The true costs and implications of V3 are discussed here: https://www.interconnects.ai/p/deepseek-v3-and-the-actual-co...

View on HN

This is great context for the cost claim. Which turns out only to be technically true when looking at the final run.

Thank you for providing this context and sourcing. I've been trying to find the root and details around the $5 million claim

Good luck, whenever an eyepopping number gains traction in the media finding the source of the claim become impossible. See finding the original paper named, "The Big Payout" that was the origin for the claim that college graduates will on average earn 1M more than those who don't go.

In this case it's actually in the DeepSeek v3 paper on page 5

https://github.com/deepseek-ai/DeepSeek-V3/blob/main/DeepSee...

Reply on news.ycombinator.com