Epoch AI estimates that LLM costs are falling by over 10x per year at fixed intelligence https://epoch.ai/data-insights/llm-inference-price-trends, so I expect prices to fall at this level of quality. I also think there's a good division of labor here where you can use the smartest model to write the script and prompts, and then use a cheaper model for all the glue work and minor revisions that trigger the tool calls.