Open-weight models are rewriting the AI pricing playbook, and it's not good news for the incumbents. Instead of giving every task to the most expensive model in town, companies are now routing routine work through cheap, open-weight systems and saving the tough stuff for the premium tier.
Customer loyalty? That's going the way of the dodo. Customer service platform Pylon scored roughly $1.6 million in free tokens from one vendor this year, plus months of unlimited usage. AI companies are buying loyalty with subsidies.
Chinese open-weight models like Kimi and GLM have started crashing the US enterprise party. Over the past two weeks, roughly half of Hex's AI data analytics platform customers plugged Kimi into their workflows. Legal AI firm Harvey, meanwhile, lets GLM-5.2 handle the grunt work, only calling in Fable 5 for the real head-scratchers.
OpenAI and Anthropic's sky-high valuations rest on two shaky assumptions: that enterprises will guzzle top-tier models and keep paying premium prices forever. Open-weight models are busy cracking both pillars.
The duopoly may still own the smartest models, but they'll be left handling the hardest, scarcest jobs. As more token consumption shifts to cheap alternatives, their revenue growth and pricing power will take a hit—and they'll keep handing out free credits and subsidies just to stay in the game.
The price war fueled by open-weight models is battering the pre-IPO valuations of OpenAI and Anthropic.