Unconfirmed: An X post describes a specific multi-model "advisor" workflow in Claude Code as dramatically cheaper, and Automatica has not confirmed either the configuration or the savings.
In a 14 November post, X user @abir35627 wrote that, in their view, "Anthropic just released the best trick for Claude Code," describing a four-model chain in which "Fable 5.2 thinks → Sonnet 5.5 builds → Opus 5.5 reviews → Haiku 5.5 coordinates a swarm of agents".@abir35627 on X1
According to the same post, the setup is presented as a general-purpose workflow for Claude Code tasks in which "Sonnet 5.5 builds, Opus 5.5 reviews on the spot, and a swarm of Haiku 5.5 agents searches for information and does research, at a per-token cost 80 times lower than Fable".@abir35627 on X1 The author characterizes the approach as "1 minute + the /advisor command = 3.4 times cheaper," without providing underlying measurement data in the visible text.@abir35627 on X1
The post further claims that in this configuration "Fable 5.2 never writes code" and "only kicks in at 3 points: before finalizing a plan, when an error keeps recurring, and before declaring the work done," with cheaper models handling all other steps.@abir35627 on X1 The rationale offered is that "You only pay Fable's price for making decisions. The cheap models do everything else."@abir35627 on X1
As a setup guide, @abir35627 tells readers to "Run claude update" and states that "Fable 5.2 requires Claude Code v2.1.321 or later" before pasting a prompt that, according to the post, "will first read Anthropic's official documentation on advisor mode."@abir35627 on X1 The post says users should configure "Sonnet as the main model, Fable as the advisor, Haiku for the subagents, and Opus as the reviewer," and asserts that the script "backs up your configuration and shows you every change before applying it, comparing before and after."@abir35627 on X1
The author also claims there is a specific configuration detail that gates the alleged benefit: "there's a line that tells Sonnet 5.5 WHEN to consult Fable. Without it, the advisor didn't activate even once in the 198 cases in Anthropic's test".@abir35627 on X1 No separate documentation of that test is linked in the excerpt.
The post states that it will provide "the full prompt and Anthropic's official documentation on advisor mode" and refers readers to an article "to get the most out of Fable 5.2," but those materials are not included in the available evidence excerpt, and Automatica has not reviewed them.@abir35627 on X1
Anthropic has not, in the material available to Automatica, announced a "best trick" for Claude Code matching this description, confirmed that this particular four-model chain reflects official guidance, or endorsed any of the claimed cost ratios. The post attributes some behavior and testing to "Anthropic" but does not link to a primary Anthropic source in the provided text.@abir35627 on X1
To substantiate or refute the circulating claim, developers would need at least three pieces of evidence: Anthropic documentation describing the same advisor-mode workflow; reproducible benchmarks demonstrating the stated "3.4 times cheaper" outcome and "80 times lower" per-token costs under controlled conditions; and independent confirmation that advisor invocations in "198 cases" match the pattern described. None of that corroboration appears in the sources Automatica reviewed.
The broader context is that orchestration choices around model selection, tools and memory have become a recognized performance lever in agent systems. Earlier coverage of Google's Harness² framework noted that "performance depends not only on the underlying model but also on the prompts, tools, memory and orchestration around it."Automatica Press2 The X post applies a similar logic to Claude Code, but its specific configuration and claimed gains currently rest on a single, self-reported account.@abir35627 on X1
— Daniel Rowe, Policy Correspondent (Policy, Law & Geopolitics)