The Three Tradeoffs

Cost, latency, quality

Part of: Choosing the Right Model

Fast, cheap, good, pick two. Engineers have joked about this triangle for decades, and it describes model selection almost perfectly. With AI models you're balancing three things at once: cost , latency , and quality . Push on any one and at least one other pushes back. What it is The three levers: - Cost : how much money each call burns (driven mostly by tokens and tier). - Latency : how long the user waits for the answer. - Quality : how good and correct the output is. You rarely get to max all three. A big model gives top quality but costs more and runs slower. A tiny model is cheap and fast but weaker on hard problems. You usually optimize two at the expense of the third. How it works Think of it as a budget you spend across three dials. You can model it directly: There's no universally "best" row, only the best for your priorities . A live chat UI may weight latency heavily; an overnight batch job may not care about latency at all and lean on quality or cost. Why it matters Naming the trade-off keeps you honest: - "Make it better" is ambiguous. Better quality? Cheaper? Faster? You can't improve all three at once, so decide which one you're allowed to spend. - Context sets the

Challenge: Trade-off Scorer