Fable as Architect, Grok 4.5 as Typist: Routing 90% of Coding Tokens to a Cheaper Model

You can now route 90% of coding tokens to Grok 4.5. A free open-source Claude Code plugin does the routing. Fable stays on as the architect. It owns specs, judgment, and verification. The implementer does the typing. The model ships at $2 per million input tokens. Output is $6 per million tokens. That undercuts Sonnet on list price for code volume. It also uses about half the tokens of peer frontier models. You can drop the expensive model from most keystrokes. Optional lanes still race a second vendor when stakes are high. One stack now covers the full loop: > Expensive model as architect > Cheap model as default typer > Optional second vendor race > Cross-vendor review built in The plugin is MIT and free. Install it and log into the CLI. Leave judgment to the top model.

I traced this to a real, specific open-source project: fable-advisor by DannyMac180.

Confirmed accurate:

• Grok 4.5 pricing checks out exactly: Grok 4.5 is priced at $2 per million input tokens and $6 per million output tokens. OpenRouter
• "Uses about half the tokens of peer frontier models" is a fair, even slightly conservative characterization: xAI reports Grok 4.5 resolves SWE-Bench Pro tasks with 15,954 output tokens on average versus 67,020 for Opus 4.8 (max) — about 4.2x fewer, and other sources describe roughly 2x token efficiency versus comparable models — "about half" is within that range. Developers Digest
• "Undercuts Sonnet on list price for code volume" holds up on output cost, which dominates coding workloads: Claude Sonnet 5 is priced at $2 per million input tokens and $10 per million output tokens through August 31, 2026, then $3/$15 after, versus Grok's flat $2/$6 — Grok's output rate is 40-60% cheaper than Sonnet's, even during Sonnet's discounted intro window, and the gap widens further once Sonnet reverts to standard pricing. Eden AI
• The plugin itself is real and matches the description almost word for word: "Implementation mechanics are ~90% of a session's tokens and Grok 4.5 handles them at near-parity — so this runs far cheaper than Fable-for-everything, and every implementation comes from a different model family than the architect that reviews it: cross-vendor review is built into the routing, not bolted on. For high-stakes work, race grok-implementer and codex-implementer on the same spec and let the architect pick the stronger diff." This maps directly onto "90% of coding tokens," "cross-vendor review built in," and "optional lanes still race a second vendor when stakes are high." GitHub
• It genuinely is free, open-source, and installed via claude plugin marketplace add + Grok CLI login, matching "Install it and log into the CLI."

Important caveat — this describes an older version of the plugin:

• The project has since moved on from the exact setup this post describes. Version 4 of fable-advisor restructures the routing: the session architect moves from Fable to Opus, the Grok 4.5 lane is removed entirely, and codex-implementer (GPT-5.6 Sol) becomes the default typing lane. If you still want the Grok lane, you have to manually pull grok-implementer.md from the older v3.1 tree. So "Fable stays on as architect" with "Grok 4.5" as the default implementer was accurate for an earlier release, but as of the plugin's current default configuration, that's no longer how it ships out of the box — a reader following "install it and log into the CLI" today would get Opus-as-architect and GPT-5.6 Sol as the default implementer instead, unless they specifically dig up the older Grok-lane file. GitHub