Model Prices Are Quietly Collapsing at the Top
Argon, Sol, and Sonnet 5.5 all sit near $2/$10. The frontier got cheaper per token even as it got smarter. That changes the math.
Look at the rate cards from this week's launches: Gemini 4 Argon at $2/$10, GPT-6.1 Sol at $2/$10, Claude Sonnet 5.5 at $2/$10. A year ago that bought you mid-tier. Now it buys the frontier. The top of the market got cheaper per token even as the models got better.
The compression matters more than the benchmarks. When the best model costs what a decent one used to, the 'is the premium worth it' calculation flips. You can run the frontier on the boring tasks without flinching, which is where most tokens actually go.
It also squeezes the mid-tier. If a $2 model does what a $10 model did last year, the $10 model has to justify the gap or drop. The cheaper decider models on top of this mean the whole stack is getting leaner, not just the flagship.
For a builder, the takeaway is to stop caching old price assumptions. Re-quote your per-call cost every quarter; the number you modeled six months ago is probably wrong, and almost certainly lower now.