Anthropic pushes Claude Opus 4.8 beyond code completion with dynamic workflows

The update brings dynamic workflows to Claude Code, adds effort controls, cuts fast mode costs, and puts more emphasis on model honesty during long-running AI coding tasks.

Anthropic has released Claude Opus 4.8 alongside dynamic workflows for Claude Code, effort controls, and lower-cost fast mode

Anthropic has released Claude Opus 4.8 alongside dynamic workflows for Claude Code, moving its flagship model further into large-scale coding work, agentic workflows, and enterprise AI software development.

The model is available now through claude.ai, Claude Code, and the Claude API. Standard pricing remains unchanged from Claude Opus 4.7 at $5 per million input tokens and $25 per million output tokens, while fast mode is now 2.5 times the speed and three times cheaper than for previous models.

The release is aimed at teams using Claude for coding, reasoning, agentic tasks, and practical knowledge work. For education, workforce skills, and technical training, it adds another marker of how AI coding tools are moving from code suggestions toward project-level software work, including migrations, refactors, bug fixes, and multi-step engineering tasks.

Dynamic workflows are launching in research preview for Claude Code users on Enterprise, Team, and Max plans. The feature allows Claude to plan work, run hundreds of parallel subagents in a single session, verify outputs, and report back to the user.

Rahul Patil, CTO at Anthropic, framed the release around trust as much as performance. Claude Opus 4.8 moved from 64.3 to 69.2 on SWE-bench Pro, but Patil pointed instead to how the model handles its own mistakes: "But the improvement I keep coming back to is honesty."

Dynamic workflows move Claude Code into larger engineering tasks

Claude Code’s dynamic workflows feature is designed for software tasks that stretch beyond a single prompt, file, or short coding exchange.

Anthropic says Claude can now plan the work, distribute it across hundreds of parallel subagents, and verify outputs before handing work back to the user. With Claude Opus 4.8, those agents can run for longer before reporting.

Patil described the target use case as "the work that used to take a quarter and a working group: codebase-scale migrations, sprawling refactors, and bug fixes across hundreds of thousands of lines, graded against the test suite you already trust."

That gives the release a more concrete enterprise and skills angle than another model score update. Codebase-scale migrations and large refactors are the kinds of tasks that sit inside enterprise software teams, university IT departments, research engineering groups, and technical training environments where learners and staff increasingly need to understand how AI coding agents work.

Anthropic says Claude Code with Opus 4.8 can carry out codebase-scale migrations across hundreds of thousands of lines of code from kickoff to merge, using an existing test suite as the bar for completion.

The research preview is available in Claude Code for Enterprise, Team, and Max.

Model honesty becomes part of the product pitch

Anthropic is also using Claude Opus 4.8 to make a more pointed claim about reliability in AI coding and agentic work.

The company says the model is around four times less likely than Claude Opus 4.7 to allow flaws in code it has written to pass without comment. The release also says early testers found Claude Opus 4.8 more reliable and sharper in judgment when performing agentic tasks.

Patil put that more bluntly in his own post: "It tells you what it's unsure of instead of dressing up thin progress as finished work."

That is where the update becomes relevant for organizations using AI agents in live workflows. A coding model that flags uncertainty or identifies its own weak work can reduce the human effort needed to inspect long-running agent outputs, although the release does not remove the need for oversight.

Patil added: "For anyone whose agents run with real oversight cost, that's worth more than another point on a leaderboard."

Anthropic says its Alignment team found Claude Opus 4.8 "reaches new highs on our measures of prosocial traits like supporting user autonomy and acting in the user’s best interest." The company also says rates of misaligned behavior, including deception or cooperation with misuse, are substantially lower than Claude Opus 4.7 and similar to Claude Mythos Preview.

The full alignment assessment and pre-deployment safety tests are included in the Claude Opus 4.8 System Card.

Effort controls and API changes give users more control

Claude Opus 4.8 also introduces effort controls on claude.ai and Cowork, giving users a choice over how much effort Claude puts into a response.

Higher effort settings are designed for deeper work. Lower effort settings return faster responses and use rate limits more slowly. Anthropic says Claude Opus 4.8 defaults to high effort, which it considers the best balance of quality and user experience.

Users can also choose extra, known as xhigh in Claude Code, or max for more difficult tasks and long-running asynchronous workflows. Patil cautioned builders that xhigh should be used deliberately: "It's strong, but it's token hungry, so reach for it deliberately."

Anthropic has increased rate limits in Claude Code to account for the higher token usage of more demanding effort settings. The Messages API has also been updated so developers can include system entries inside the messages array, allowing instructions to be updated mid-task without breaking prompt cache or routing the change through a user turn.

Claude Opus 4.8 is available everywhere today. Developers can use claude-opus-4-8 through the Claude API, with dynamic workflows available in research preview for Claude Code Enterprise, Team, and Max users.

Anthropic is also continuing to test Claude Mythos Preview under Project Glasswing with a small number of organizations using it for cybersecurity work. Anthropic says Mythos-class models need stronger cyber safeguards before general release and expects to bring those models to customers in the coming weeks.