Sakana's Fugu router outperforms Fable 5 through model selection alone

Sakana doesn't disclose which models actually feed the router or how Fable 5 was tested in isolation. The real claim is that orchestration logic outperforms a frontier model Sakana chose not to include, which inverts the burden of proof: we're asked to trust the absence of a competitor as evidence of superiority.

This announcement arrives as federal AI funding accelerates sharply (the Trump administration's $5 billion Genesis Mission grants announced this week), which tilts incentives toward inference efficiency and cost arbitrage over raw model scale. Sakana's pitch is economically timely but strategically convenient: if routing beats frontier models, smaller labs and startups funded under that new federal umbrella have a narrative justifying investment in orchestration layers rather than model training. The timing suggests Sakana is positioning itself for that wave, not proving a technical breakthrough.

If Sakana publishes results using the same benchmark on a locked test set (not their own eval) that includes Fable 5 in the routing pool, and the gains persist above 5 points, the claim moves from marketing to credible. Until then, watch whether any independent lab (Hugging Face, OpenAI, or academic group) reproduces the v1.1 results; absence of reproduction within 60 days signals the benchmark may not travel.