The 2023-2026 model price war — and why your cost model is already obsolete (step 8/8) · production tradeoffs

Checkpoint

One last thing before we move on. pass this to mark the lesson done, or skip and keep moving. hop to the next when you're ready.

Final drill. The pattern Perplexity and Cursor built into their products from day one: routing logic that picks the cheapest model that meets a quality floor under a monthly budget cap.

Build pick_model(constraint) that takes a constraint dict:

max_cost_per_month: float — the budget cap in dollars per user per month
quality_floor: str — one of "haiku", "sonnet", "opus" (the model must be at LEAST this good)

And returns one of: "haiku", "sonnet", "opus", or "none" if no model meets both the floor AND the budget.

Quality ranking (low to high): haiku < sonnet < opus. A quality_floor of "sonnet" means sonnet or opus qualifies; haiku does not. Among models that meet the floor, pick the cheapest one that stays under budget.

Use this fixed usage profile (don't change it):

20 messages/day
500 avg input tokens, 200 avg output tokens
30 days/month
prices: haiku $1/$5, sonnet $2/$10, opus $5/$25 per 1M tokens

Five constraints run for you. Expected output:

tight budget, haiku floor:   haiku
mid budget, sonnet floor:    sonnet
big budget, opus floor:      opus
tight budget, sonnet floor:  none
big budget, haiku floor:     haiku