Anthropic has released Claude Opus 4.8, making the updated artificial intelligence model available at the same price as its predecessor.
The launch adds new user controls in claude.ai, a research preview of dynamic workflows in Claude Code, and lower pricing for fast mode on Opus 4.8. Fast mode runs at 2.5 times the speed and costs one-third as much as previous models, according to Anthropic.
Claude Opus 4.8 succeeds Opus 4.7 and is positioned as an upgrade for coding, reasoning, agentic work and practical knowledge tasks. Anthropic says the new version is designed to act as a more effective collaborator, and early testers found it more reliable on agentic tasks.
A central claim in the rollout is improved honesty. Anthropic says AI models can sometimes overstate progress or reach conclusions without enough evidence, and that Opus 4.8 is more likely to flag uncertainty and less likely to make unsupported claims.
Internal evaluations found Opus 4.8 was about four times less likely than Opus 4.7 to let flaws in code it had written pass without comment, Anthropic says. Its alignment team also found lower rates of behaviour such as deception or cooperation with misuse than in Opus 4.7, with performance similar to Claude Mythos Preview on those measures.
New controls
One of the main product changes is an effort control in claude.ai and Cowork. It lets users choose how much effort the model puts into a response, with higher settings aimed at deeper reasoning and lower settings aimed at faster replies and slower use of rate limits.
The control is available on all plans. Opus 4.8 defaults to a high-effort setting, which Anthropic says offers the best balance between output quality and user experience.
For coding tasks, Anthropic says that default effort level uses a similar number of tokens to the default setting in Opus 4.7 while delivering better performance. Users can also choose extra or max effort settings, which consume more tokens in exchange for better results, particularly on difficult tasks and long-running asynchronous workflows.
Anthropic has increased rate limits in Claude Code to reflect the higher token use at these settings. The change is intended to give users more flexibility in choosing a level suited to their projects.
Dynamic workflows
Anthropic has also introduced dynamic workflows in research preview for Claude Code. The feature allows Claude to plan work, run hundreds of parallel subagents in a single session, and verify outputs before returning them to the user.
The feature is designed for larger software tasks, including codebase-wide migrations across hundreds of thousands of lines of code. Anthropic says those workflows can run from kickoff to merge while being checked against an existing test suite.
Dynamic workflows are available in Claude Code for Enterprise, Team and Max plans. The addition reflects a broader industry push to move AI tools from chatbot-style interaction toward longer-running task execution inside software development environments.
Anthropic has also changed the Messages API so developers can place system entries inside the messages array. That means instructions can be updated mid-task without breaking prompt caching or routing the change through a user turn.
Developers could use that approach to change permissions, token budgets or environmental context while an agent is running, Anthropic says. The update is likely to matter most to developers building multi-step agent systems that need tighter operational control during execution.
Pricing and access
Regular pricing for Opus 4.8 remains unchanged from Opus 4.7 at USD $5 per million input tokens and USD $25 per million output tokens, Anthropic says. Fast mode is priced at USD $10 per million input tokens and USD $50 per million output tokens.
Developers can access the model through the Claude API under the name claude-opus-4-8. Opus 4.8 is available everywhere Claude is offered, according to Anthropic.
Anthropic also pointed to work on lower-cost models that offer some of the same functions as Opus. A small number of organisations are already using Claude Mythos Preview for cybersecurity work, it says, but models at that level need stronger cyber safeguards before a broader release.
Anthropic says its Alignment team concluded that Opus 4.8 “reaches new highs on our measures of prosocial traits like supporting user autonomy and acting in the user's best interest.”