When the cheap model is the right model
Frontier models earn their price on hard problems. Planning a system around a difficult room. Weighing tradeoffs across ten components. Anything with many variables and no clean answer.
Most questions aren’t that. What DAC stands for, typical planar headphone impedance, what a crossover does. A budget model answers these just as well, for a fraction of the credits.
A simple rule: lookups, definitions, and quick checks go to a fast, cheap model. Analysis, planning, and multi-component decisions go to a frontier model. And if a cheap answer feels thin, re-ask the same question on a stronger model. Escalating when needed costs far less than defaulting to expensive for everything.
Go deeper
Sorting questions by what they need
Most hi-fi questions fall into one of three shapes. Lookups: what a term means, what a typical value is, what a component does. Checks: is this combination safe, does this number make sense, did I read the spec sheet right. Decisions: which of these five should I buy, how should I plan this room, what do I upgrade first given everything else.
Lookups and checks have short, well-established answers that appear in the training data many times over. A small model handles them cleanly. Decisions have many variables, no single right answer, and depend on your specific situation. That is where a frontier model, especially one that reasons, pays for itself.
Practical takeaway: before you pick a model, ask which of the three shapes your question has. The shape decides the model.
What a small model gets right
"What is a Class D amplifier?" "Is 87 dB sensitivity low?" "What does DSD stand for?" "Convert 6 ohms to what an amp rated at 4 and 8 ohms would see." A fast model answers all of these correctly, in seconds, for a fraction of the credits a frontier model would spend on the same words. It also reads the Pure Neo database the same way the big models do, so a lookup of a specific product’s specs is grounded regardless of which model you chose.
Where the small model shows its limits is in juggling. Ask it to compare three amplifiers against two speakers in a specific room and it may drop a constraint halfway through, or give each pairing a fair summary and no overall verdict.
Practical takeaway: for single-fact questions and definitions, the small model is the right model, and the grounding is identical.
Escalation beats defaulting
The expensive mistake is setting a frontier reasoning model as your default and leaving it there. Every "what does this mean" then costs the same as a system design session. The cheap mistake, in the other direction, is asking a small model a big question and trusting a thin answer because it sounded fine.
The habit that avoids both: start small. If the answer comes back vague, drops something you said, or contradicts itself, switch models in the same chat and ask again. You have now paid for one cheap answer and one strong one, which is still less than defaulting to strong for everything.
Practical takeaway: pick the small model first. Escalate when the answer earns it, and only then.
A worked week
Monday, a lookup: what tracking force your cartridge wants. Small model. Tuesday, a check: whether your tonearm’s effective mass suits that cartridge’s compliance. Small model, with the database behind it. Wednesday, a decision: replace the cartridge or the phono stage first, with a budget and a stated preference for vocals. Frontier model, reasoning on. Thursday, a follow-up definition that came out of Wednesday’s answer. Back to the small model in a fresh chat.
Four questions, one expensive. That ratio is what a monthly credit allowance is sized for.
Practical takeaway: most weeks look like this. One decision, several lookups. Spend accordingly.