Anthropic launches Claude Opus 4.8 with new agent features
Anthropic has released Claude Opus 4.8, an updated model that the company says improves on Opus 4.7 across coding, reasoning, and agent tasks. The launch also adds dynamic workflows in Claude Code, effort controls in claude.ai, and lower fast-mode pricing for Opus 4.8.
Anthropic has released Claude Opus 4.8, a new version of its flagship model that the company says improves on Opus 4.7 across benchmarks and real-world agent tasks. The model is available now at the same regular price as Opus 4.7, while fast mode is now cheaper for this version.
According to Anthropic, Opus 4.8 is a more effective collaborator and performs better on coding, reasoning, computer-use, browser-agent, and practical knowledge work tests. The company said the model is more reliable when carrying out multi-step tasks, and early testers reported better judgment, fewer unsupported claims, and stronger consistency during long sessions.
⚡ New to this?
This is a model update from Anthropic, the company behind Claude. A model is the AI system doing the work, and better benchmark scores usually mean it is more capable at tasks like coding, research, and using tools on a computer.
The new features matter because they change how people can run AI assistants in practice: more control over how hard the model thinks, better support for large multi-step jobs, and lower costs for faster responses. For teams using AI in software, legal, or data workflows, those changes can affect both reliability and cost.
🦞 OpenClaw angle
If you run self-hosted or agentic workflows, test Opus 4.8 on your hardest multi-step jobs first: migrations, tool-using agents, and long research runs. Use the new effort controls to separate fast, cheap tasks from high-stakes ones, and set higher effort only where the extra token use is justified.
If your agent stack depends on changing instructions mid-run, update your harness to use Messages API system entries so you can adjust permissions, budgets, or context without restarting the session. Also re-check your validation logic: Anthropic says Opus 4.8 is better at surfacing uncertainty, so your automation should treat those signals as a reason to verify outputs before merging or acting on them.
Anthropic said Opus 4.8 delivered the highest score it has recorded on its Legal Agent Benchmark and became the first model to break 10% overall on the all-pass standard. The company also said it is the strongest computer-use and browser-agent model it has tested, with an 84% score on Online-Mind2Web. In addition, Anthropic said Opus 4.8 is around four times less likely than its predecessor to let flaws in code it writes go unremarked.
The company highlighted several customer and partner reactions in the release. Testers from products including Claude Code, Cursor, Devin, CoCounsel Legal, Databricks’ Genie, and Hebbia said the model showed better judgment, more efficient tool use, stronger citation precision, and better handling of long, complex tasks.
Anthropic also said Opus 4.8 is more honest in the sense that it is less likely to claim progress without evidence and more likely to flag uncertainty. The company’s alignment team said the model reached new highs on measures such as supporting user autonomy and acting in the user’s best interest, while showing lower rates of misaligned behavior than Opus 4.7.
The release arrives with several new product features. In Claude Code, a new research-preview feature called dynamic workflows lets Claude plan large tasks, run hundreds of parallel subagents in one session, and verify outputs before reporting back. Anthropic said this can support codebase-scale migrations across hundreds of thousands of lines of code, with the existing test suite used as the standard for validation.
On claude.ai and Cowork, users now have an effort control next to the model selector. Anthropic said higher effort settings make Claude think more deeply, while lower effort settings return answers faster and use rate limits more slowly. The company said the control is available on all plans.
Anthropic also updated the Messages API so system entries can now be placed inside the messages array. The company said developers can use that to update Claude’s instructions mid-task without breaking the prompt cache or routing the change through a user turn. Anthropic said this can help with changing permissions, token budgets, or environment context while an agent is running.
Opus 4.8 defaults to high effort, which Anthropic said offers the best balance of quality and user experience. For harder jobs and long-running workflows, the company recommends using extra effort or max effort. Anthropic said it raised Claude Code rate limits to accommodate the higher token use at those settings.
Claude Opus 4.8 is available now through Claude API and in Claude products. Regular pricing stays at $5 per million input tokens and $25 per million output tokens, while fast mode is priced at $10 per million input tokens and $50 per million output tokens.
