Introducing Claude Opus 4.5

Source: Anthropic Date: 2025-11-24 URL: https://www.anthropic.com/news/claude-opus-4-5

Summary

Anthropic launched Claude Opus 4.5 on November 24, 2025, at $5/$25 per million input/output tokens. Claims: “best model in the world for coding, agents, and computer use”; state-of-the-art on SWE-bench Verified; top on SWE-bench Multilingual (7 of 8 languages); top on BrowseComp-Plus agentic search. New features: “effort parameter” for API users to balance capability and cost, up to 76% fewer tokens at medium effort, expanded context management. Cited as “most robustly aligned model” with improved prompt injection resistance. API identifier: claude-opus-4-5-20251101. Early partners: GitHub Copilot, Cursor, Notion.

Implications

• Claude model cadence / Claude Code thread. Opus 4.5 in November 2025 is the first full-generation Opus at the “4.x” naming, distinct from the Sonnet cadence. The $5/$25 pricing puts it above the Sonnet tier and establishes a clear three-tier structure (Haiku/Sonnet/Opus) at $1/$3/$5 input.
• “Effort parameter.” The ability to tune capability vs. cost per request is significant for enterprise customers who need predictable cost control for high-volume agentic workflows. This makes Opus 4.5 viable for production at scale, not just premium use cases.
• 76% token efficiency at medium effort. If this holds in production, it substantially changes the unit economics of Opus-tier deployments — an enterprise customer previously forced into Sonnet for cost reasons can potentially use Opus at Sonnet cost.
• SWE-bench Multilingual (7/8 languages). Moving beyond English coding benchmarks is a signal about international developer market targeting — and a benchmark that open-source multilingual coding models will target.
• Watch: whether the “effort parameter” becomes the dominant way enterprises deploy Opus; Cursor/GitHub Copilot adoption rates for Opus 4.5 vs. Sonnet; prompt injection resistance in adversarial testing post-launch.