Analysis
CNBC ran a piece last week calling it "model fatigue" -- labs rolling out version after version as the gap between them compresses. I think that framing gets the direction backward. Fatigue implies slowing down. What actually happened in the 72 hours ending this issue: Sakana AI split Fugu into two tiers and cut pricing 40-60%, DeepSeek shipped a new Flash model that reportedly beats its own flagship on select benchmarks, and OpenAI opened its internal agent harness to every developer. That's not fatigue. That's three labs of very different sizes independently deciding the fastest way to compete is to ship again, immediately, rather than wait for the next planned release cycle.
What most people are missing: this compresses faster for smaller labs than big ones, which inverts who actually benefits. Sakana AI is a fraction of OpenAI's size, and it just matched frontier labs on price competitiveness within one release cycle by specializing instead of trying to out-scale them. If that pattern holds, the labs racing to build ever-bigger frontier models are competing against a treadmill that smaller, faster-moving labs can now hop onto with a fraction of the compute budget.
“What most people are missing: this compresses faster for smaller labs than big ones, which inverts who actually benefits.”
For founders building on any single model, the diligence item is concrete: check how many major re-pricings or model swaps your primary vendor has pushed in the last 90 days, not the last year. If it's more than two, your unit-economics model is stale by definition, and your board deck should stop presenting inference cost as a fixed input.
Room for disagreement: it's entirely possible this pace is itself unsustainable -- training and serving costs still have to be paid somewhere, and a lab that ships three pricing tiers in a year without commensurate revenue growth is burning cash faster than the headlines suggest. CNBC's fatigue framing may end up right for a different reason than intended: not that users are tired of new models, but that labs eventually run out of runway to keep shipping them this fast.
