Perplexity's $0.04 Decider Quietly Changed What 'Cheap' Means
A 27B model that decides for a fraction of a cent turns 'should I call an LLM' into a question you can ask a thousand times.
Perplexity open-sourced pplx-decider-v1-27b at $0.04 per million input tokens with free output. That price isn't a discount, it's a category. For a yes-or-no routing decision, calling a frontier model at $2 is overkill; calling this is a rounding error, so you stop rationing the decision.
The use I actually found: a tiny model that decides which bigger model should handle a request. Route the easy stuff to a cheap model, the hard stuff to a frontier one, and your average cost drops without your users noticing. The decider pays for itself because it sits in front of every call.
Cloudflare's Clef and Clef-flash, also open and cheap with free output, point the same way. The interesting layer of 2026 isn't the biggest model, it's the smallest one smart enough to say 'send this elsewhere.' Decision models are infra, not features.
If you're building anything with an LLM in the loop, a $0.04 decider is the cheapest optimization you haven't tried. It turns 'do I really need the big model for this' from a design meeting into a line of code.