Anthropic has launched Claude Opus 5, positioning it as the Opus model for everyday agentic coding and professional work rather than only a premium research curiosity.
The July 24 release says Opus 5 is available today, is the new default model on Claude Max, and is the strongest model on Claude Pro. Developers can use claude-opus-5 on the Claude API at $5 per million input tokens and $25 per million output tokens, the same base price Anthropic lists for Opus 4.8.
That pricing and placement are the important product facts. Anthropic is saying more of its long-running agent capability can move into normal Opus use cases, with effort settings for trading off capability, speed, and token use.
Anthropic is selling steadier agent work
The company frames Opus 5 around coding, knowledge work, scientific research, visual outputs, and multi-step execution. Its launch page says the model comes close to Claude Fable 5 frontier intelligence at half the price, while remaining behind Mythos 5 on cybersecurity tasks.
Those benchmark claims should be read as Anthropic’s claims unless an independent evaluator reproduces them. Still, the product direction is clear. Opus 5 is meant for work where a model has to inspect its own progress, use tools, revise outputs, and finish a task without creating extra operational drag.
GitHub’s same-day Copilot changelog reinforces that positioning. GitHub says Claude Opus 5 is rolling out for Copilot Pro+, Max, Business, and Enterprise users across Visual Studio Code, Visual Studio, Copilot CLI, Copilot cloud agent, the Copilot app, github.com, mobile, JetBrains, Xcode, and Eclipse. Business and Enterprise admins must enable the model policy before organization users can select it.
Safeguards are part of the product shape
Anthropic says Opus 5 has stronger safeguards for a narrow range of cyber tasks and that flagged Claude.ai, Claude Code, and Claude Cowork requests fall back to Opus 4.8 by default. The API also gets beta support for automatic fallbacks and mid-conversation tool changes.
That tells buyers where the operational complexity lives. A more capable model is not simply an upgrade toggle. Teams need to know what happens when a request crosses a safety boundary, whether the fallback model is acceptable for the task, and how model effort affects cost and reliability.
For developers, the practical checkpoint is migration behavior. Anthropic says disabling thinking on Opus 5 is allowed only at high effort or below; trying to disable thinking at xhigh or max returns an error. That is the kind of detail that can break an automated workflow if a team treats the new model as a drop-in replacement.
