Matthew Berman recommends pairing Fable 5 as an advisor with GPT-5.5 to cut agent token costs
A small X conversation about model routing has turned into a practical debate over whether the cheapest useful agent is the one that asks a stronger model for help only when it has to.
1964.1K2195K512.2K
“Use Fable + GPT-5.5 for peak performance and lowest quota usage.”
Matthew Berman@MatthewBermanTECH#1784Story Brief
In a post on X, Matthew Berman argued that developers should "use Fable + GPT-5.5 for peak performance and lowest quota usage," describing a workflow where a stronger model stays in an advisor role while a cheaper model does most of the work. In a second high-engagement post, Peter Steinberger on X pushed a similar idea in a slightly different direction, recommending Codex as the workhorse and linking a GitHub skill file for his version of the setup.
“If you run this workflow, ask Fable to make codex the workhorse.”
Peter Steinberger@steipeteRelated Stories
2 linkedPeter Steinberger's Codex-first version of the workflowShows how Steinberger describes a Codex-heavy variant of the setup he linked from his X post.8h agoGitHubMatthew Berman's original X postContains the core claim that pairing Fable with GPT-5.5 gives high performance with lower quota usage.9h agoX
Matthew Berman's original X postContains the core claim that pairing Fable with GPT-5.5 gives high performance with lower quota usage.9h agoX
