Nine in ten business uses of AI can be handled well by many models
Glean picks the model for each task to hold down costs, but it is too new to know if customers will accept Chinese-built models.
20VC with Harry Stebbings · Why OpenAI and Anthropic Won't Win the App Layer | Glean FounderAbout 90% or more of business uses of artificial intelligence can now be handled well by many models.
Arvind Jain, founder of Glean, which makes AI software that connects to a company's documents and helps employees find answers, agreed that shows commoditization of the model layer.
He said Glean holds down customer costs by choosing the model for each task, turning to an open-source model, one anyone can download and run, when it should give a strong answer, but only if the customer accepts open models. He said it is too early to know how many will.
That choice only recently became possible, he said, with a recent Chinese model the first his own team trusts for most work. The open question, he said, is whether companies will accept a Chinese-built model, because of fears about hidden risks and because competitors could use it against them if its use became known.
Questions people ask
What is an open-source AI model?
It is a model whose underlying code and weights are publicly released, so anyone can download, run, and modify it. Well-known examples include Meta's Llama family and China's DeepSeek and Qwen models. Because there is no per-use fee to a model vendor, running them is often much cheaper, though you still pay for computers to run them.
How can using a cheaper model still give a good answer?
Most office tasks, like summarizing documents, answering policy questions, or drafting routine replies, do not need the most advanced reasoning. A smaller or open model can often produce an answer that is just as useful. Software can therefore send easy tasks to cheap models and reserve expensive frontier models for hard tasks.
Why would a company prefer one model over another if most are good enough?
Cost is the biggest reason, since AI bills can run into millions of dollars a year. Companies also weigh speed, reliability, and where data is processed. Some also avoid certain models over concerns about security, legal terms, or the country where the model was built.