Anthropic Launches Claude Opus 4.8 With New Effort Controls

Anthropic has officially released Claude Opus 4.8, the latest version of its flagship AI model, which aims to provide more reliable and collaborative results [1]. The update introduces new features designed to improve model transparency and performance, including adjustable effort settings and a research preview for dynamic workflows [1].

Key takeaways

• Claude Opus 4.8 is designed to be more honest and is four times less likely to report flaws in its own code compared to previous versions [1].
• Users can now manually adjust the model's "Effort" settings—Low, Medium, High, or Max—to balance response speed against reasoning depth [1].
• A new "Dynamic Workflows" feature in research preview allows the model to coordinate hundreds of subagents to complete complex tasks [1].
• The "fast mode" for Opus 4.8 is now three times cheaper than it was for previous models [2].
• Anthropic plans to release a more advanced "Mythos-class" model in the coming weeks [1].

Enhancing Collaboration and Control

Anthropic claims that Opus 4.8 is more inclined to support user autonomy and is significantly less likely to be deceptive or provide unverified information [1]. Early testers, including engineers at Shopify, have noted that the model demonstrates improved judgment, often asking clarifying questions and catching its own mistakes rather than simply agreeing with user prompts [2]. This shift is intended to move the AI away from being a "yes-man" and toward acting as a more critical, effective collaborator [3].

To manage how the model processes information, Anthropic has introduced an "Effort" selector in the model interface [1]. While lower settings prioritize speed and reduced token usage, higher settings encourage the model to think longer and provide more detailed responses [1]. For the most complex tasks, the company recommends using the "Max" effort mode [1]. Additionally, the new "Dynamic Workflows" feature allows the model to plan and execute tasks by running hundreds of subagents in parallel, which the company says can handle massive codebase migrations [3].

Why it matters

The release of Opus 4.8 reflects a broader industry trend toward making AI models more "agentic"—capable of autonomous reasoning and verifying their own work before presenting results [3]. By focusing on honesty and the ability to flag uncertainty, Anthropic is attempting to address common criticisms regarding AI hallucinations and reliability [3]. While some users have expressed skepticism regarding benchmark improvements, the company is already looking ahead to the next phase of its development [2]. Anthropic has confirmed it is working on a new class of models, currently referred to as "Mythos," which are expected to offer even higher intelligence in the coming weeks [1].