Anthropic releases Sonnett 5.5 priced at USD 2/mln input tokens, USD 10/mln output tokens, USD 0.20/mln cache reads

The relevant transmission into listed equities runs through the proxy set: hyperscalers and AI-exposed software names on the inference-cost deflation theme, and the accelerator complex on whether cheaper per-token pricing signals efficiency gains that compress compute demand or elasticity that expands it. Episodes of this kind have historically produced a two-sided read, with model-level price compression treated as margin pressure for the labs themselves and as demand stimulation downstream, and the market has tended to resolve the ambiguity by watching the peer response rather than the launch itself. The cache-read pricing tier is the structural tell worth noting, since steep discounts for cached context reflect the shift toward agentic and long-context workloads where input costs dominate, and rivals have typically matched such structures within a short window. Anthropic's prior form is rapid iteration with aggressive pricing, so the follow-ons are whether competing labs reprice their flagship tiers and how the hyperscalers that both fund and distribute these models frame unit economics on their next earnings calls.