que paso
IT Brew reports that companies including SmartBear, Tanium and Smartsheet are experimenting with open-weight AI models as the cost of using proprietary models increases. Executives cited potential savings, lower latency, local hosting and reduced dependence on major providers, while emphasizing that most experiments have not yet become large-scale deployments.
IT Brew reports that Fitz Nowlan, SmartBear’s vice president of AI and architecture, is conducting an extensive evaluation of open-weight models after seeing Chinese models narrow the performance gap with offerings from major AI labs. Nowlan told the outlet that SmartBear is considering routing some prompts to open-weight systems if the cost-benefit analysis is favorable. The company is testing whether these models can support valuable technical use cases while reducing operating costs and latency. Taken together, this account describes evaluation at the level of specific use cases and operating choices. It does not describe a completed migration, and the decision remains tied to the companies’ own assessments of performance, cost and fit.
According to IT Brew, Tanium is also evaluating open-weight models. Tanium CTO Harman Kaur said the company must determine which model to use, how to host it and which applications should be tested first. The report says Tanium has found that some tasks perform comparably to closed models, but still faces questions about governance and whether Chinese models should be provided to customers. Kaur is evaluating the Inkling model from Thinking Machines Lab. That evaluation leaves several decisions open at once. The report’s examples connect model choice with hosting, application selection and customer use, so a favorable result in one task would not by itself settle the broader deployment question.
IT Brew also reports that Smartsheet currently relies mainly on Anthropic and Amazon models while exploring open-weight alternatives. CEO Rajeev Singh said the company could eventually route simpler tasks to cheaper models, but is prioritizing speed and time to market for now. The article quotes PwC Global Chief AI Engineer Scott Likens as saying open-weight models are a major client topic, although he does not see adoption at scale. These accounts are reported by IT Brew and are not independently confirmed in the supplied source. The examples therefore show interest without establishing a common adoption pattern. Their significance lies in the range of questions being considered, while the supplied reporting continues to distinguish experimentation from confirmed production use.
Lea la fuente principal: itbrew.com ↗
Por qué es importante
The report describes a practical shift from asking what AI can do to asking what it costs to operate at scale. Open-weight models could give companies more control over hosting and model selection, but they also create additional responsibilities around evaluation, customization, security, governance and possible bias.
The report frames open-weight models as a cost and control question rather than simply a technical contest. Companies that have moved beyond early prototypes may now have enough user traffic for inference charges to affect unit economics. IT Brew quotes Iz Beltagy, director of AI research at the Allen Institute for AI, saying that businesses with significant traffic and user interaction data have stronger reasons to consider local or self-managed models.
Open-weight systems may offer flexibility that proprietary application programming interfaces do not. Businesses can potentially host them in the cloud or on their own infrastructure, select different models for different tasks and avoid relying entirely on one provider. The article says executives are particularly interested in lower latency, cost efficiency and protection against the possibility that a model provider could develop a competing software product. Those benefits remain prospective for the companies described.
The trade-off is that downloading model weights does not eliminate the cost or risk of deployment. IT Brew reports that organizations may need to pay for infrastructure, conduct rigorous evaluations, fine-tune or customize models and investigate model provenance to reduce security risks. The article also raises concerns about bias associated with model creators and about whether technical performance on tasks such as tool calling is sufficient to address broader governance questions. The source provides no independent cost comparison, security audit, benchmark methodology or production reliability data.
Qué ver a continuación
The key indicators will be whether experiments move into production, which tasks companies assign to open-weight models, and whether lower inference costs offset setup and maintenance expenses. Companies will also have to decide whether models developed by Chinese firms are acceptable for customer-facing or regulated uses.
The immediate question is whether the companies cited by IT Brew move beyond evaluation. SmartBear’s routing plans, Tanium’s model and governance review, and Smartsheet’s possible use of open-weight systems for simpler tasks are all described as exploratory. The source does not provide deployment dates, adoption volumes, measured savings, error rates or evidence that any of the experiments have reached customer-facing production.
Model selection will be another important signal. The article identifies Chinese systems such as Z.ai’s GLM-5 and DeepSeek v4 as examples of models attracting interest, while also mentioning open-weight offerings from U.S. companies. Companies will need to assess not only benchmark performance but also licensing, update practices, training-data provenance, security exposure, censorship or bias concerns and the handling of sensitive information. IT Brew reports these concerns but does not independently verify them.
Finally, watch whether the economics change as usage grows. The report suggests that companies may keep expensive frontier models for complex work while sending routine requests to cheaper systems, creating a multi-model strategy. That approach could lower average costs, but it may add routing, monitoring and quality-control overhead. The supplied source does not establish that open-weight models will materially reduce total costs, displace proprietary providers or become standard across enterprise AI.
