Agentic Coding: How Task Routing Cut My AI Coding Bill by 90%

Описание и применение

Сокращение стоимости разработки на 90 % за счёт маршрутизации задач к самым дешевым моделям, без использования frontier‑моделей.

Задача

High cost of using frontier LLMs for all coding tasks, especially when many steps do not require high reasoning.

Результат

Zero frontier tokens, 12–18× cost reduction, same functional outcome as a frontier‑only run.

Рабочий процесс

• Plan task with planner model (Nemotron Ultra) via MCP tool
• Implement task with implementer model (Nemotron Super) via MCP tool
• Review changes with reviewer model (DeepSeek Flash) via MCP tool
• Optionally summarize logs with summarizer model (Nemotron Super) via MCP tool

Доказательства

$0.205 vs. $2.41–$3.62, an 11.8×–17.6× cost difference.

Источник

Zero tokens went to Claude or any other frontier provider, anywhere in the loop.

Источник