Anthropic Claude Opus 4.5
Anthropic positioned Claude Opus 4.5 as its most intelligent model and 'the best model in the world for coding, agents, and computer use,' also citing meaningful gains on everyday work like deep research, slides, and spreadsheets. It claimed state-of-the-art software-engineering results (top scores on SWE-bench Verified) and notable token efficiency — at medium 'effort' it reportedly matches the prior Sonnet 4.5's best SWE-bench score using 76% fewer output tokens, via a new API 'effort' parameter (low/medium/high). It arrived as the Opus-tier successor in the Claude 4 line, following Sonnet 4.5 on September 29, 2025.
Opus 4.5 shipped the same day on Anthropic's apps and API and across all three major clouds (AWS, Google Cloud, Microsoft), with a sharply lower price of $5/$25 per million input/output tokens — roughly a third of prior Opus pricing — plus Claude Code upgrades like an improved Plan Mode.
On launch evidence Opus 4.5 looks like a strong, incremental consolidation of Anthropic's coding/agentic lead rather than a paradigm shift — but it is recent enough (announced late November 2025) that a settled, independent verdict on real-world reliability and durability is still forming, so treat the benchmark and efficiency claims as vendor-reported until third-party use accumulates. The most consequential move is arguably commercial: cutting Opus-tier pricing to $5/$25 brings frontier capability to a far broader set of teams and pressures competitors on cost-per-task, not just raw capability. The 'effort' control reflects a maturing emphasis on token efficiency and predictable spend, which matters more to enterprise budgets than leaderboard wins. Executives should watch whether early benchmark leadership holds under sustained agentic workloads before treating it as a standard.
It signals the frontier competing on cost and efficiency, not just capability — Opus-level performance at roughly a third of the old price changes the economics of deploying AI agents at scale.