Despite the hype around large AI models, many companies like Meta are using small models for routine tasks, finding them more practical and cost-effective
For many tasks in corporate America, it's not the biggest and smartest AI models, but the smaller, more simplistic ones that are winning the day X: @wsj . LinkedIn: Christopher Mims and Max Keenan Bluesky: @carlquintanilla and @jessefelder X: @wsj : The kinds of sophisticated AI models that companies are using to get real work done and reduce head count aren't the ones getting all the attention, writes Christopher @mims https://www.wsj.com/... LinkedIn: Christopher Mims : This week for The Wall Street Journal I wrote about a small thing in AI with big implications: — Small models — not large ones … Max Keenan : Great article from Christopher Mims laying out some of the nuance of applying AI at scale and how AI isn't just magic … Bluesky: Carl Quintanilla / @carlquintanilla : MIMS: The “unsung heroes of AI, the ones actually transforming business processes and workforces, also happen to be the smallest, fastest and cheapest.” — @mims.bsky.social 👀 — www.wsj.com/tech/ai/larg... [image] Jesse Felder / @jessefelder : “The reality is, for many of the operations that we need computing for today, we don't need large language models.” www.wsj.com/tech/ai/larg...
Context & Ripple Effects
Earlier coverage showed startups using AI to raise output with smaller teams, while the cost of deploying advanced reasoning models can rise even as token prices fall. This report supplies the enterprise counterpart: routine work is being matched to smaller models rather than to the most capable systems.
The economics also fit the prior push to build capable small models cheaply through distillation techniques. Meta’s use signals that model selection is increasingly an operational decision about task fit, speed, and cost—not only benchmark capability.
First-order effects
- Companies handling routine operations can shift more workloads to smaller, faster, cheaper models, reducing the cost and latency of those deployments.
- For Meta and similar adopters, AI implementation becomes more practical for narrowly defined workflows; the reported use is also tied to efforts to reduce head count.
Second-order effects
- Providers of frontier models face stronger pressure to demonstrate a lower cost per completed task, especially where reasoning-model token consumption has raised developers’ bills.
- Enterprise buyers gain more leverage to use a portfolio of models by workload, rather than standardizing on the largest available model.
Third-order effects
- If this pattern persists, enterprise AI spending may split between premium models for exceptional tasks and efficient smaller models for high-volume routine work.
- The competitive advantage may shift from owning the largest model toward integrating, operating, and measuring the right model reliably across business processes.
The trend: Enterprise AI is moving from a race for maximum model capability toward workload-specific optimization of cost, speed, and business output.