Some ideas for what comes next, May 2026
The practical performance gap between closed and open AI models is widening, as real-world agentic usefulness—not just benchmarks—becomes the critical measure. While open models continue advancing technically, their lack of equivalent "agent moments" and the absence of major competitors to products like Claude Code indicate they will likely specialize in areas like enterprise automation rather than disrupting the core knowledge work dominated by closed labs, which is key to future AI revenue.
Deep Analysis
Background
The article frames 2026 as a year of continuous, escalating AI disruption without pause. It critiques the common discourse on the open-closed model gap, arguing that the real litmus test is not benchmark scores but real-world utility in agentic harnesses—software environments where models act as autonomous assistants for complex tasks like coding. The "Opus 4.5 in Claude Code moment" of December 2025 is cited as the definitive example of a closed model achieving transformative utility.
Key Points
- The Agent Utility Gap is the True Metric: The time delay between open and closed models is best measured by whether open models can replicate the agentic robustness of the best closed models. The author predicts this gap will take 12+ months to close, if it closes at all, making products like Claude Code and Codex appear as distinct, superior categories.
- Evidence from Industry Leaders: The fact that Google's Gemini lacks a clear competitor to Claude Code and Codex is presented as strong evidence for the gap's persistence. The author suggests Gemini's strengths may align better with Google's existing products (search, YouTube) rather than being a general-purpose agentic workhorse.
- A Specialized Future for Open Models: The article predicts open models will not have a breakthrough, general-capability "god model" like Mythos in the near term. Instead, due to resource constraints and strategic choices, open model labs will specialize. They are expected to dominate in "automated, enterprise agents and low-cost domains" rather than competing directly with leading closed labs (OpenAI, Anthropic, Google) in cutting-edge knowledge work tools.
- Economic Feedback Loop: This specialization is self-reinforcing. The high-value, high-revenue market is currently driven by agentic tools for knowledge workers. Closed labs' dominance here fuels their economic engine to fund future models, while open labs, focused on other domains, may fall further behind in the race for general frontier capabilities.
Significance
The analysis challenges the narrative that open models are closing in on closed models by highlighting a practical and economic divergence. The significance lies in the prediction that the AI landscape will not converge but bifurcate: closed models as the engines of premium, transformative AI for individuals and knowledge work, and open models as the workhorses for specialized, cost-effective, automated systems in enterprise and niche sectors. This shapes expectations for innovation, competition, and where future AI impact and revenue will be concentrated.
Disclaimer: The above content is generated by AI and is for reference only.