Cutting AI Coding Costs by Routing Most Work to Commodity Models
Melvin Vivas · X post · 2026-08-27 · Open on X
Topics: AI Dev Tools & Productivity, LLMOps, Deployment & Monitoring · Level: intermediate
Summary
A quoted case study: a lead engineer used the Pi coding-agent harness to send most of his client work to DeepSeek. He used a stronger model (Fable) only when he needed it. Using cheaper 'commodity models' by default kept his total usage within two $20 plans.
Key points
- Send everyday coding tasks to cheap commodity models (here DeepSeek).
- Use the more expensive model (Fable) only for tasks that need it.
- With this setup, two $20/month plans covered all of his usage.
- The harness (Pi) was called 'the most important piece' of the setup because it makes switching models easy.
Resources mentioned
- Pi · tool · x.com · free
A customizable coding agent that can be extended through its Extensions API and connected to several model providers.
Also in: 100-Iteration Autonomous Coding Loop with Local Qwen on an RTX 3090 (Melvin Vivas on X · notes), Pi Agent: A Highly Customizable Agent with Durable Sessions (Melvin Vivas on X · notes), Prompting an Autonomous Agent Loop: Be Specific, Show Visible Progress (Melvin Vivas on X · notes), Running an Overnight 100-Iteration Agent Loop with a Local Qwen Model (Melvin Vivas on X · notes) and 54 more - DeepSeek · tool · deepseek.com · free
DeepSeek's updated Flash model, said to be faster with shorter outputs, available through DeepSeek's web app and API.
Also in: DeepSeek-V4.1-Flash-1017 release: better benchmarks, shorter outputs (Melvin Vivas on X · notes), Open models now dominate token volume on Vercel AI Gateway (Melvin Vivas on X · notes), jina-ocr-v1: Turning PDFs, Scans and Tables into Markdown (Melvin Vivas on X · notes), Bolt.new Adds Open Models (GLM, DeepSeek, Kimi) via Bolt Forge (Melvin Vivas on X · notes) and 5 more - Fable · tool · anthropic.com · paid
Named as what the creator used with Devin for this build; the post gives no details about what it is.
Also in: Is AI Getting Cheaper? How to Measure LLM Cost Per Task and Per Intelligence Level (Melvin Vivas on X · notes), Demo: One-Shotting a Flappy Bird iPhone App with Devin and Fable 5.1 (Melvin Vivas on X · notes), Subagents with mixed models: a strong planner and a fast executor (Melvin Vivas on X · notes), Cognition's SWE-2 Coding Model Now in Devin (Melvin Vivas on X · notes) and 11 more
More in AI Dev Tools & Productivity
- herdr: Run Pi, Codex and Claude Code in One Terminal
- Cursor Adds App Creation, Code Storage with Origin, and Vercel Deploys
- Herdr Crash Course: A Terminal Multiplexer for Running Many Coding Agents
- Using Codex (GPT 5.6 Luna High) for Dataset and Evals Work
- Fork Open-Source Apps and Add Missing Features with a Coding Agent
- Let Claude Code or Codex handle Git commits, pushes and releases