Anthropic launches Claude Opus 4.5 with major upgrades in coding and reasoning

Anthropic has officially launched Claude Opus 4.5, which a major upgrade to its flaghip AI model that pushes deeper into real software engineering, research, and multi-step workflows. The model is now available across the Claude apps, API, and major cloud platforms, with pricing set at $5/$25 per million tokens.

Anthropic says Opus 4.5 offers the best coding performance of any frontier model today. It leads on SWE-bench Verified and even outscored every human candidate on Anthropic’s own two-hour engineering exam. Early testers have also reported that the model now handles ambiguity better, reasons through trade-offs, and cracks multi-system bugs with fewer prompts.

Well, it’s not just coding where Claude Opus 4.5 delivers. Improvements ranges across reasoning, math, multilingual tasks, and vision understanding, placing it ahead of earlier models on several benchmark suites. The model also shows sharper agentic behavior, often finding valid but unexpected paths through constrained tasks.

Anthropic isn’t compromising on safety either, as it says Claude Opus 4.5 its most aligned and most robust model yet, showing higher resistance to strong prompt-injection attacks. The company highlights progress across “concerning behavior” evaluations and broader stress testing.

Developers also get more control through a new effort parameter, which lets teams choose between fast, efficient replies or deeper, more deliberate reasoning. At medium effort, Opus 4.5 matches Sonnet 4.5 on SWE-bench with far fewer tokens. At peak effort, it surpasses Sonnet 4.5 while using under half the output tokens.

Anthropic is rolling out product upgrades as well. Claude Code gets a refined Plan Mode and desktop support. Meanwhile, Claude for Chrome is expanding to Max users, and Claude for Excel now reaches Max, Team, and Enterprise customers. Long chats in the Claude apps also benefit from automatic context summarization to keep conversations flowing.