source: techcrunch ai: can tech companies learn to love cheaper ai models?
level: business
the ai industry has long assumed that bigger models are better, but mounting costs are forcing a rethink. coinbase co-founder brian armstrong predicts that within 12 to 18 months, 80% of workloads will run on models that are 99% cheaper, while only 20% will need the latest, most powerful systems. this shift could reshape the economics of ai, as companies discover that many tasks don't require top-tier models.
early tests support this idea. legal ai startup harvey, working with fireworks ai, cut inference costs by three times without sacrificing quality by routing simpler tasks to a cheaper model and reserving a more advanced one for complex work. harvey co-founder gabe pereyra noted that quality now means using the most efficient model for each job, not just the most powerful one. this approach challenges the industry's scaling-first mindset, which has been fueled by investor subsidies.
the real competition is between large and small models, not proprietary and open ones. users can save money by switching to smaller versions from the same labs or to open-weight alternatives. as token prices rise and subsidies fade, enterprises face cost pressure for the first time. they might respond by reducing usage or abandoning less promising projects, but if most tasks can be handled by smaller models, it could dampen demand for expensive frontier training and hurt labs like openai and anthropic ahead of their ipos.
why it matters: if cheaper models can handle most tasks, it could reduce the need for expensive frontier models, impacting ai company revenues and the pace of advanced model development.
source: techcrunch ai: can tech companies learn to love cheaper ai models?