SaaStr cuts app costs 72% switching models, signals margin reset for ANZ SaaS
## The Numbers SaaStr Connect, the company's AI matching and recruiting app, cut infrastructure costs by 72% after moving to Haiku 5.5 and a cheaper model mix. Same workflow, lower bill. Jason Lemkin, SaaStr founder, says it was the first time he saw costs fall in production since early 2025, despite token price drops across the board. The implication: if you set your model mix at launch and have not touched it since, you are probably overpaying. Your competitors who re-benchmarked are underpricing you with better margins. ## The Pricing Pressure Lemkin attributes Anthropic's quarterly growth deceleration to pricing, not demand. OpenAI is selling comparable products at roughly 30% lower prices. Open-weight models are pulling prices down further. When the category leader hits a ceiling on premium pricing, every B2B vendor priced as a markup on that capability hits the same ceiling. For sales teams selling SaaS or AI products: if most of your value is model capability, plan for your price to fall 30 to 40% within a year. Shift pricing toward the outcome you deliver and the workflow you own, because those do not deflate the way tokens do. ## The Agent Cost Reality Persistent consumer agents cost more than $20 per user per month to serve. Power users cost far more. Replit and Lovable have tuned their infrastructure down to roughly $4 to $5 per user per month for on-demand use, but always-on agents run north of $20. Elon capped free usage of his agent product. Amjad says the era of free agent subsidies is ending. For sales tech buyers: measure cost-to-serve separately for your top 10% of users before you set a budget. Blended averages hide power users who cost 5 to 10 times what they pay. ## The ANZ Angle Lower AI delivery costs create room for Australian and New Zealand SaaS vendors to expand margins or compete on price. The decision is which lever to pull. The DoorDash–Instacart transaction (approximately $17 billion, pending regulatory approval) demonstrates how platform scale can reshape adjacent local-commerce markets. DoorDash generated about $4.6 billion in Q3 revenue and completed nearly 1 billion orders. Instacart reported approximately $1.1 billion in revenue and more than 90 million orders in the latest comparable quarter. Worth noting: re-benchmark your production workflows every time a new small model ships, using your real tasks, not public evals. A 70% drop in inference costs changes your gross margin enough to justify the work.