Anthropic has released Claude Opus 4.5, calling it its “best model for coding, agents, and computer use.” The model is now available across Anthropic’s apps, API, and major cloud platforms, with pricing set at $5/$25 per million tokens.
Developers can access the model through the Claude API using the ID claude-opus-4-5-20251101, the company said.
According to the company, Opus 4.5 delivers stronger real-world software engineering performance.
On SWE-bench Verified, Anthropic said the model scored highest among frontier systems. Compared to Google’s recent launch of Gemini 3 Pro, it scored 80.9%, while the latter scored 76.2%.
Early internal testers reported more reliable task execution. “Opus 4.5 just gets it,” Anthropic said, summarising feedback from colleagues who found the model more capable of handling ambiguity and multi-step debugging. Customers with early access shared similar responses.
Anthropic also benchmarked Opus 4.5 on its internal performance engineering take-home exam. The company said the model scored higher than any human candidate ever within the prescribed 2-hour limit, while noting that the test measures only technical skills under time pressure, not collaboration or long-term judgment.
The model showed gains across coding tasks, multilingual programming, and reasoning datasets. It also found unconventional but valid solutions in agentic benchmarks such as τ2-bench. It scored 88.9%, while Gemini 3 Pro scored 85.3%.
Anthropic highlighted safety as a major update. In its system card, the company said Opus 4.5 is a robustly aligned model. It claims significant improvement against strong prompt-injection attacks. “Opus 4.5 is harder to trick with prompt injection than any other frontier model in the industry,” Anthropic said.
Developers will see new controls through the Claude API, including an effort parameter that lets users trade off speed and capability. At medium effort, Anthropic said Opus 4.5 matches Sonnet 4.5’s best SWE-bench performance “while using 76% fewer output tokens.”
The company is also expanding product integrations. Claude Code gains an upgraded Plan Mode and is now available in the desktop app, enabling multiple parallel sessions. Claude for Chrome is rolling out to all Max users, and access to Claude for Excel—announced in October—is expanding to Max, Team, and Enterprise.
For users already on Opus tiers, Anthropic has removed model-specific caps and raised usage limits. “We’re updating limits to make sure you’re able to use Opus 4.5 for daily work,” the company said.
Anthropic noted that Opus 4.5’s improvements in context management and tool use also strengthen multi-agent workflows and long-running research tasks. As larger shifts in the nature of software work emerge, the company said it plans to share further findings through its Societal Impacts and Economic Futures research.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.