The Small Model Won My Workflow
The frontier models get the headlines, but the small, cheap, fast model quietly took over my daily work. For summarization and extraction, it's all I need.
The frontier models get the headlines, but the small, cheap, fast model quietly took over my daily work. For summarization, extraction, and classification, it's all I need, and it's a tenth of the cost.
The September 2026 price war made this obvious. OpenAI's GPT-6 Luna at ten cents per million tokens and similar budget tiers from DeepSeek and GLM mean the cheap model is now good enough for most bulk work, and the savings are real at volume.
I reserve the expensive frontier model for the few tasks that genuinely need it, hard reasoning, tricky code, and let the small one handle the hundred small jobs. The split cut my bill without cutting quality where it counts.
The lesson isn't 'cheap is best.' It's 'match the model to the task.' Paying frontier prices for a summary is like hiring a surgeon to bandage a cut.
If your AI bill is high, audit what actually needs the big model. You'll probably find most of it doesn't, and the small model was waiting the whole time.