Erik Bradley from ETR, now a Futurum company, presented findings on the "model wars" in enterprise AI, focusing on how actual enterprises are rating and using various models rather than just benchmark rulings. ETR's research spans 16 years, drawing insights from over 10,000 vetted technology leaders overseeing significant annual technology spend. This proprietary dataset, comprising over 67,000 responses, is proven to correlate with vendor performance and offers forward-looking data across global verticals and industries, initially serving Wall Street and now providing deep market intelligence.

Regarding frontier models, the data indicates a dynamic competitive landscape. Anthropic Claude has shown massive growth, rising from 50% to almost 75% adoption among enterprises building their own AI applications from March to September. Claude leads in five of six tracked use cases, including complex tasks like coding and software development, and boasts the highest overall satisfaction among users. However, it is also noted as the least cost-effective, frequently running over budget. Microsoft Copilot, while remaining broadly used, is seeing its lead against Claude shrink, with some users even replacing it. Google Gemini has remained relatively flat in adoption, and Grok shows minimal presence in the enterprise space.

The presentation also highlighted the growing impact of open weight models. The share of AI token usage going to open weight models rose from 23% to 34% in 12 months, with expectations to exceed 40% soon. While 93% of enterprises are running open weights alongside proprietary models for shared work, only 7% are replacing foundational models outright. Open weights are primarily used for routine, low-complexity, and cost-sensitive tasks. Google Gemma is showing significant growth among open weight models. Enterprises are actively managing AI consumption by shifting to cheaper models or tiers and trying to reduce tokens per task, with a substantial portion even increasing spending despite nearly 50% of respondents not yet seeing a measurable return on investment or not actively tracking it.

Recorded live in Santa Clara, California on October 14, 2026 as part of AI Field Day 9. Watch the full presentation at https://techfieldday.com/appearance/etr-presents-at-ai-field-day-9/ or visit https://etr.ai or https://techfieldday.com/event/aifd9/ for more information.