Computerworld: open models gain traction

Computerworld reported on May 19 that enterprise IT teams are increasingly trying smaller, open AI models alongside proprietary systems from OpenAI and Google. The publication said buyers are drawn by lower running costs, more control over deployment and better fit for narrow business tasks. Gartner analyst Deepak Seth told Computerworld that open models give companies a starting point they can customize without building a model from scratch. The article adds to a broader 2026 debate over whether enterprise AI demand will keep concentrating around the biggest closed models or spread across cheaper, more specialized alternatives. ### Which companies and models are being pulled into this shift? Computerworld named Meta’s Llama, Mistral, DeepSeek and Minimax among the open models gaining traction with enterprise buyers. It also said proprietary AI providers have released open-source or open-weight offshoots, including Google’s Gemma, OpenAI’s GPT-OSS and Microsoft’s Phi. Those products give corporate buyers more choices than a single-vendor stack built only around frontier proprietary models. (computerworld.com) ServiceNow, Microsoft, HubSpot and RWS were cited by Computerworld as companies arguing that open models are easier to blend into existing AI infrastructure, reduce computing costs and fit agentic AI workflows. Max Goss, a senior research director at Gartner, told the publication that outages at Anthropic and OpenAI are also pushing CIOs to think about vendor lock-in and resiliency. (computerworld.com) ### Why are enterprise buyers looking at open models now? Deepak Seth told Computerworld that open models help IT leaders manage the economics and governance of AI inside their organizations. Jesse Williams, co-founder and COO of Jozu, said more use cases are emerging and that open source is more flexible than proprietary systems in some settings. Computerworld said that flexibility matters for companies that want tighter control over how models are deployed internally. (computerworld.com) MIT Sloan published a separate report on January 20 saying closed models still account for about 80% of model usage on OpenRouter, but open models reach roughly 90% of closed-model performance at release and then narrow the gap. The same report said closed models cost users, on average, six times as much as open ones, and that shifting demand toward open models could save the global AI economy about $25 billion annually. Frank Nagle, a research scientist at MIT’s Initiative on the Digital Economy, said organizations should use “the right tool for the right job” instead of defaulting to the most popular model. (computerworld.com) ### Are open models replacing the biggest proprietary systems? Computerworld said no. Jesse Williams told the publication that proprietary models are still gaining usage rapidly and that open models’ rise should not be read simply as a backlash against large language models. The article instead describes a market in which enterprises are adding open models for certain workloads while keeping closed models for others. (mitsloan.mit.edu) Max Leaming, head of data science and AI solutions at ManpowerGroup, told Computerworld that open models are trained on lower volumes of data than top proprietary systems and may not be as capable across broad tasks. He said companies need to experiment to determine what each model is good at, adding that “none of them are truly general purpose models.” ### What does this look like inside an enterprise deployment? (computerworld.com) IBM CEO Arvind Krishna said at the company’s Think 2025 conference that smaller, domain-specific models are faster, cheaper and easier to place where companies want them to run. He said such models can be as much as 30 times less expensive to run than conventional large language models, and pointed to IBM’s Granite family as an example of smaller open-source systems aimed at business tasks. IBM’s framing matches the argument in Computerworld’s May 19 report: enterprises are not choosing only the largest model available, but matching model size and openness to cost, speed, security and integration needs. ### What should readers watch next? May 19 is unlikely to settle the open-versus-closed debate, because enterprise adoption still depends on production results rather than benchmark claims. Computerworld’s reporting suggests the next evidence will come from vendor deployments, CIO purchasing patterns and how companies balance resiliency, governance and cost across multiple model providers. Gartner, Jozu, IBM and large software vendors are among the named participants likely to surface those next data points as 2026 enterprise rollouts continue. (computerworld.com 1) (computerworld.com 2)