OpenAI’s Ted Sanders shares a rule of thumb for picking the right GPT-model
What changed
OpenAI’s Ted Sanders explained how GPT-6.1 Terra fits within OpenAI’s model lineup. Terra is a smaller GPT model that, despite its reduced size, outperforms the larger Astra model in technical communication tasks. It also comes with significantly lower operational costs. However, Astra still leads in most other capabilities beyond technical writing.
Sanders emphasized that evaluating GPT models requires looking at at least three separate dimensions: model size, reasoning effort, and the complexity of multi-agent systems. He argued against ranking models by a single “IQ score” because these dimensions affect performance and costs differently.
Why builders should care
Choosing the right GPT model often means balancing cost, performance, and application focus. Terra’s edge in technical communication at a fraction of Astra’s cost makes it an efficient option for builders needing precise, clear outputs without large compute expenses. But for more general or challenging tasks, the bigger Astra model still holds the advantage.
Understanding model size, the reasoning pathway, and agent collaboration as separate factors reframes how operators assess AI capabilities. This breaks the simplistic “bigger always means better” narrative, encouraging strategic choices based on specific workload needs rather than raw scale or a single score.
The practical takeaway
Operators can cut AI expenses by selecting smaller models like Terra for clearly defined, technical tasks. At the same time, they should not expect these smaller models to replace larger ones for more complex or creative workloads. The breakthrough insight is that AI performance can be decomposed across multiple dimensions, allowing smarter allocation of resources and architectural decisions.
For teams working with multi-agent AI setups, Richardson’s framework hints at tailoring agents by dimension rather than just scale, which can improve system efficiency and output relevance. This fresh perspective on model choice lets builders optimize practical trade-offs between cost, speed, and quality.
What to watch next
Look for OpenAI and competitors to release clearer guidance and tooling around selecting models based on multi-dimensional criteria, rather than flat benchmarks. More AI vendors may start offering smaller, task-specific models that deliver better cost-performance for focused applications.
Pay attention to how multi-agent systems evolve. If the separation between model size, reasoning, and agent interaction grows sharper, new design patterns and API integrations will emerge. This could reshape how AI systems are engineered for everything from automated technical writing to complex decision-making.
AI Quick Briefs Editorial Desk