Overview
- Opus 4.5 is now live across Anthropic’s apps, API, and AWS, Google Cloud, and Azure, becoming the default model for Pro, Max, and Enterprise tiers.
- Anthropic cites state-of-the-art results on coding and agent benchmarks, including 80.9% on SWE-bench Verified, and says the model outscored all human candidates on its internal engineering exam.
- API pricing drops to $5 per million input tokens and $25 per million output tokens, alongside a new “effort” parameter that lets developers trade speed and cost for depth.
- Product updates include Claude Code on the desktop app, expanded Claude for Chrome and Excel access, longer-running agent support, and automatic conversation summarization for extended chats.
- Safety documentation reports improved resistance to prompt injection yet mixed outcomes in agentic settings, with lower refusal rates for malicious requests in Claude Code and computer-use tests, and independent validation still pending.