Synopsis: A lower token price does not settle the decision to adopt an open-weight model. Enterprises still have to evaluate its behavior, establish acceptable controls and make it work with the tools employees already use. Benchmark scores may not predict the cost of completing a real task, either. Taking advantage of a wider choice of models requires evaluation and deployment infrastructure, not simply a cheaper option on a pricing page.
Ameya Kanitkar, co-founder and CTO of Larridin, joins Mike Vizard to examine why open-weight models remain a limited share of the enterprise usage his company observes. He identifies several obstacles, including uncertainty about model security, gaps between benchmark results and practical performance, and the difficulty of switching models underneath established development tools. Hosting a model in a preferred location addresses part of the deployment question, but does not by itself establish that its behavior is suitable for a sensitive workload.
Kanitkar recommends testing models against an organization’s own tasks rather than relying on leaderboard positions. He describes one evaluation in which a lower-priced model still saved money, but by much less than its advertised pricing initially suggested. His broader point is to measure the cost of achieving the required result. He also says agentic workloads dominate token spending in Larridin’s customer observations, making those workflows an important place to investigate model choices rather than assuming everyday chat usage explains the bill.
His infrastructure recommendation is an intelligent routing layer that can apply policies and change the model behind a workflow without repeatedly asking employees to learn another tool. That flexibility still depends on evaluating which models are appropriate for each task. Open weights can support customization and deployment control, but Kanitkar notes that organizations have limited capacity to vet a rapidly changing field. The practical opportunity is to combine model choice with evaluation, governance and routing, so that lower prices can translate into savings without leaving suitability unexamined.
