In 6 months, we'll likely have a Fable 5 quality model at a third of the cost. In another ~12 months, Opus 4.8 intelligence will probably be available to run locally on your own machine. Think about what that actually means: The models that cost $15 per million output tokens today will soon cost only $5. Plus, the intelligence that currently requires Anthropic's infrastructure will fit on a consumer laptop. This isn't just my speculation; it's the current AI cost and progression curve playing out in real time. GPT-3 launched in 2020 at $60 per million tokens. The same capability costs fractions of a cent today (Llama 3 runs locally on a MacBook and matches GPT-3.5 performance from two years ago). Every 12 months, the same intelligence gets roughly 3x cheaper. And every 18 to 24 months, the latest frontier models become available locally. This right here is what gets me most excited about the future of AI. And AI doomers have the nerve to say progression has slowed lol.(Miles Deutscher)