How are enterprises trying to control token spend?

This week, Ramp opened its internal LLM router to the public and said it had reduced its own model costs by 30%. Meanwhile, Moonshot launched Kimi K3, expanding the set of open-weight models enterprises can evaluate as alternatives to proprietary APIs.

Together, they reflect a broader push for more control over model selection, deployment, and cost. Qualitate’s buyer discussions suggest the infrastructure and governance required to do so are still early.

We asked Qualitate’s Research Analyst to analyze 1,500+ discussions and surface how enterprises are responding to token spend.

What tactics are enterprises using to reduce LLM costs?

• Routing tasks to cheaper models instead of defaulting to frontier ones
• Prompt caching and templating to cut token waste
• Setting  token budgets and usage caps
• + 3 more in the full briefing 

One expert put the stakes plainly:

"We had a budget and blew through it 3 times as fast as we anticipated."

–  Partner, Professional Services, Large Enterprise

For investors, the findings offer a demand-side view into several of the questions shaping the AI trade: how durable model-provider pricing power may be, whether open-weight alternatives are becoming credible, and how quickly enterprises can bring inference costs under control.

Get the briefing

Download the full Research Analyst output to see which cost-control tactics are showing up most often, where governance remains immature, and how enterprises are approaching open-source and self-hosted models,backed by direct buyer  quotes. 

The briefing is a snapshot of the current market. Qualitate customers can use Research Analyst to explore vendor-specific implications, compare buyer behavior over time, and run follow-up channel checks as new discussions enter the platform.

‍