No pricing changes this week across the 12 tracked LLM APIs. Providers typically adjust rates quarterly at most, so extended quiet periods are standard.
The spread between cheapest and most expensive tracked models is now 50x on input ($0.2 vs $10 per million tokens) and 100x on output ($0.5 vs $50). That's wide enough that workload architecture matters: if you're running high-volume classification or extraction where Grok 4.1 performs adequately, you'd need Claude Fable 5 to deliver measurably better outcomes on essentially every request to justify the cost difference. For most production use cases, that threshold makes the expensive tier viable only when output quality directly impacts revenue or when you're processing relatively small volumes where the absolute cost difference remains negligible.
Next pricing update likely won't arrive until at least one provider announces a new model tier or adjusts rates to match competitor positioning.