The Llama models tend to spark a race-to-the-bottom in terms of pricing - I fully expect Llama 3.1 to end up cheaper than GPT-4o mini after a bunch of providers spin up endpoints that compete on price.