Anthropic has introduced Claude Opus 4.8, the latest version of its flagship Claude Opus model, saying it delivers improvements across benchmarks, stronger judgment in agentic tasks, and better collaboration for users. The new model is available immediately at the same pricing as Claude Opus 4.7, with standard usage set at $5 per million input tokens and $25 per million output tokens. Fast mode pricing is also lower than before, with Opus 4.8 able to run at 2.5 times speed for $10 per million input tokens and $50 per million output tokens.
Alongside the model release, Anthropic is rolling out several product updates. On claude.ai, users can now control how much effort Claude applies to a task, giving them a choice between faster responses and deeper reasoning. The company says higher effort settings improve response quality, while lower effort settings reduce latency and use fewer rate limits. The feature is available across all plans.
Claude Code is also gaining a new “dynamic workflows” feature in research preview. Anthropic says the tool allows Claude to break down very large software tasks, coordinate hundreds of parallel subagents in one session, and verify outputs before returning results. The company says this can support codebase-scale migrations across hundreds of thousands of lines of code, with existing test suites used as the validation standard. Dynamic workflows is available for Claude Code Enterprise, Team, and Max users.
Anthropic highlighted performance gains for Opus 4.8 in coding, reasoning, legal work, computer-use, browser automation, and enterprise data workflows. Early testers said the model shows better judgment, asks more relevant questions, catches mistakes more reliably, and stays more consistent over long sessions. The company said Opus 4.8 scored especially well on several internal and partner benchmarks, including agentic and legal evaluations, and that it is more efficient in tool use and more reliable in end-to-end task completion than earlier Opus versions.
A major emphasis in the launch is honesty and reliability. Anthropic said Opus 4.8 is less likely to make unsupported claims, more likely to flag uncertainty, and around four times less likely than its predecessor to let flaws in code go unnoticed. The company also said its alignment review found the model stronger in supporting user autonomy and acting in the user’s best interest, with lower rates of misaligned behavior such as deception or cooperation with misuse.
For developers, Anthropic says the Messages API now accepts system entries inside the messages array, allowing instructions to change mid-task without breaking prompt caching or routing changes through a user turn. The company says this can help with agent permissions, token budgets, and environment context during long-running workflows.
Anthropic said Opus 4.8 is a modest but meaningful step forward, while also signaling more ambitious plans ahead. The company said it is working on lower-cost versions of Opus-level capability and on a higher-intelligence model class under Project Glasswing. Claude Opus 4.8 is now available through claude.ai and the Claude API.
