OpenAI Executive Outlines Multi-Dimensional Approach to Model Selection

Ted Sanders of OpenAI explains that model selection should consider multiple dimensions—size, reasoning capability, and multi-agent performance—rather than relying on a single metric. He notes that smaller models like GPT-6.1 Terra can outperform larger ones in specific tasks such as technical communication while being significantly more cost-effective. Performance cannot be reduced to a simple 'IQ score,' emphasizing the need for nuanced evaluation.
GPT-6.1 Terra exemplifies a shift toward task-specific optimization, demonstrating that smaller models can surpass larger predecessors in niche domains like technical communication due to refined training and architecture. Sanders’ emphasis on multi-dimensional evaluation challenges the industry’s reliance on model size as a proxy for intelligence, highlighting that performance varies significantly across different capabilities such as reasoning, creativity, and multi-agent coordination.
This approach could influence how enterprises select AI tools, potentially lowering costs for specialized applications while encouraging more nuanced benchmarking. Developers and organizations may prioritize efficiency and domain fit over raw scale, which may democratize access to high-performing models and shape future research toward balanced, application-driven design.