Comment by ActivePattern

You would be surprised how much of a difference the model makes for certain niche tasks.

For my use case, `gemini-3.1-flash-lite` is ~20% higher accuracy than the next best model of comparable cost (considering both proprietary and open-weight alternatives)