Plan with a Strong Model, Execute with a Cheaper One

Melvin Vivas · X post · 2026-08-13 · Open on X

Topics: AI Dev Tools & Productivity, LLMOps, Deployment & Monitoring · Level: intermediate

Summary

The creator suggests a split workflow: use Fable for planning and Grok 4.6 for execution. The quoted post says Grok 4.6 followed Fable's plan with only a couple of nudges and finished the same work in 1h 24m. It used 8.6M tokens for about $55, roughly 1/10 the cost of the Fable-only run.

Key points

• Split coding-agent work into planning (strong model) and execution (cheaper model).
• Grok 4.6 followed Fable's plan with only a couple of nudges.
• Result: 1h 24m, 8.6M tokens, about $55 at per-token pricing.
• That's about 10x cheaper than doing the whole job with Fable.

Resources mentioned

• 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: 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), GPT-6 Astra Access Across Plans vs Fable on Claude Max (Melvin Vivas on X · notes) and 10 more
• Grok 4.6 · tool · x.ai · paid
  xAI's latest frontier LLM, announced as an improvement over Grok 4.5 at the same price.
  Also in: Creator's Top 3 Closed Models: Fable 5, GPT 5.6 Sol, Grok 4.6 (Melvin Vivas on X · notes), Hermes (Grok 4.6) vs ChatGPT Work on a text-to-PDF task (Melvin Vivas on X · notes), Grok 4.6 release: better than Grok 4.5 at the same price (Melvin Vivas on X · notes)

Try this

• Try having a frontier model write the plan and a cheaper model carry it out, and compare cost and quality.
• Benchmark planner/executor model pairs on one coding task, tracking tokens, time and cost.