Recommendations for Choosing Models for Agent Development
Sebastian Raschka has formulated recommendations based on OpenAI data for selecting models in agent development, which can significantly reduce costs.
Following the announcement of GPT-5.6, several models and levels of effort emerged, raising the logical question: which combinations make sense to choose for agent development?
To clarify, Sebastian Raschka took official data from OpenAI, correlated it with levels of effort, and built his own diagram, based on which he formulated practical recommendations:
If you do not need the maximum performance of Terra Ultra, it is almost always more beneficial to use the Luna model with a higher level of effort. The performance will be the same or higher, and the cost will be lower.
Models below Sol High can be disregarded — in these scenarios, it is better to use Luna with an increased level of effort.
Instead of Sol Extra High, it is better to choose Terra Ultra right away.
Additional costs for Sol Ultra are unlikely to justify the gain compared to Sol Max.
upd: in the lower left corner, there was an error - the labels for Luna are shifted by one position. Thus, Luna Ultra should be read as Luna Max.
Why it matters
AnalysisThese recommendations will help developers optimize their model selection for agent development, reducing costs and increasing project efficiency.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
GPT Image Generative Models Prompting Guide
OpenAI's gpt-image generative models are designed to produce production-quality visuals and fully controllable creative workflows. They are suitable for both professional design ta...
JSON — Structured Data for Programs, APIs, and AI Agents
A complete hands-on guide to JSON: syntax specifications, serialization via parse and stringify, REST API consumption, JSON Schema validation, and structured outputs for LLMs.
Four Resume Tips from OpenAI’s Head of Applied AI Development
Wulfie Bain, OpenAI’s head of applied AI development, shared four resume rules: keep it short, remove irrelevant details, highlight outcomes, and lead with the strongest achievements.
MIT researchers show GPT-6 Astra can improve other AI models by editing their weights
In an experiment, Astra raised Qwen3.5-0.8B's accuracy on MATH-500 from 7.9% to 48.1% by changing only 0.016% of the model's weights.