Anthropic cut Opus cache-read prices 60%.

Cheaper inference now matters more than benchmark wins for developers in lower-income markets.

OpenAI halved GPT-6 Sol’s price as both labs face margin pressure from 20-gigawatt datacenter spending, pushing toward price cuts over capability leaps.

Anthropic said Opus 5.5 only modestly beats OpenAI’s new GPT-6 Astra model in coding benchmarks.

Whether cost overtakes capability as the main axis of competition among frontier AI labs remains unresolved.

How each outlet framed it
Ars Technica
quantifies 40% cost savings over Opus 5 and explains automatic safety-guardrail routing for high-risk tasks

Sources: Ars Technica