For agentic coding, one can say: - Unless you need Terra Ultra perf, it's always better to use a Luna model with higher effort setting (same or better performance but cheaper). - Forget everything below Sol High, use Luna with higher effort settings here - Forget Sol Extra High, use Terra Ultra here - The extra cost of Sol Ultra is probably not worth it over Max

Sorry, there's been a label shift in the original plot I made above. The corrected plot is below. I.e., Ultra = Max and so on. The relative comparisons between models is still correct. (Afaik Ultra runs Max with 4 subagents mainly for speedups) The numbers come from the Artificial Analysis Coding Agent Index as mentioned in the chart. And here's the source link: artificialanalysis.ai/agents/coding-…

@rasbt Poor little gemini, all alone at the bottom of the graph...

@AndroidBlogger I think they were planning to release some better models this summer (based on what I remember from the I/O keynote)

@AndroidBlogger @rasbt gemini 3.1 pro preview is on par with the smallest model luna with low level reflexion... that says a lot about the gemini model performance i guess :D