Claude Opus 4.5 (www.anthropic.com)

Anthropic today launched Claude Opus 4.5, a new frontier model it says is optimized for coding, agentic workflows and “computer use,” and is available via apps, the Claude API (claude-opus-4-5-20251101) and major clouds. Pricing is announced at $5/$25 per million tokens, positioning Opus-level performance as more affordable for teams. Alongside the model release are platform updates—longer conversations, Excel/Chrome/desktop integrations, and developer tools like an “effort” parameter to trade speed/cost for capability, plus improvements to Claude Code (Plan Mode) and multi-agent orchestration. Technically, Opus 4.5 focuses on token efficiency, sustained reasoning, and robustness: Anthropic reports it uses far fewer tokens (half or more in many tests, up to 65% less) while improving pass rates, beating Sonnet 4.5 and competitors on internal benchmarks (Terminal Bench +15%). Notable gains include better long-horizon coding, autonomous multi-step workflows, Excel automation (+20% accuracy, +15% efficiency), large drops in tool-calling and build/lint errors (50–75%), faster self-improvement for agents (peak in ~4 iterations), and stronger prompt-injection resistance. The combination of improved context/memory, the effort control, and multi-agent coordination signals a practical step toward cost-effective, reliable agentic systems for software engineering and complex enterprise automation.