Artificial intelligence company Anthropic has unveiled Claude Opus 4.8, a major update to its flagship AI model that brings new controls for users, enhanced coding capabilities, and improvements in reasoning and agentic tasks.
The new model is now available through claude.ai, Claude Code, and the Claude API under the identifier "claude-opus-4-8."
Claude Opus 4.8 keeps existing pricing structure
Anthropic said Claude Opus 4.8 retains the same pricing as its predecessor, Claude Opus 4.7.
The standard version costs $5 per million input tokens and $25 per million output tokens. Meanwhile, the fast mode is priced at $10 per million input tokens and $50 per million output tokens.
According to the company, fast mode operates at approximately 2.5 times the speed of the standard version, offering quicker responses for users who require faster processing.
One of the most notable additions is a new effort control feature available on claude.ai and Cowork.
Users can now choose how much computational effort Claude invests in generating a response. Higher effort levels consume more tokens but can improve output quality for complex tasks.
Claude Opus 4.8 defaults to a high-effort setting. Anthropic said coding tasks at this level consume roughly the same number of tokens as Opus 4.7 while delivering stronger results.
For more demanding workloads, users can select an "xhigh" setting, which allocates additional computing resources. To support the increased token usage, Anthropic has also raised Claude Code rate limits.
The company said these controls align with its broader move toward token-based pricing rather than fixed subscription tiers.
Claude Code gets dynamic workflows
Anthropic has also introduced dynamic workflows to Claude Code, allowing the AI assistant to handle more complex software development tasks.
The feature enables Claude to plan projects, launch parallel sub-agents, verify outputs, and generate final reports.
According to Anthropic, dynamic workflows are designed for large-scale software projects and can work with codebases containing hundreds of thousands of lines of code.
Currently in research preview, the feature is available to Enterprise, Team, and Max subscribers.
The company has also upgraded its Messages API with live update capabilities.
Developers can now modify instructions, permissions, token limits, and contextual information while an AI agent is actively working on a task.
Anthropic said these updates can be made without interrupting prompt caching or requiring a new user interaction, making long-running tasks more flexible and efficient.
Improvements in coding, reasoning, and safety
Anthropic claims Claude Opus 4.8 outperforms Opus 4.7 across several areas, including coding, reasoning, knowledge work, and agentic workflows.
The company said the new model is approximately four times less likely to approve flawed code without identifying potential issues.
On the safety front, Anthropic reported lower levels of deceptive behavior and a reduced tendency to comply with misuse attempts compared with Claude Opus 4.7.
The company added that Opus 4.8's safety profile is similar to that of Claude Mythos Preview.
External testing involved organizations working in software development, legal services, finance, and research.
According to CursorBench, Claude Opus 4.8 completed tasks using fewer tool interactions while maintaining comparable output quality. Another tester reportedly found the model's cost-performance ratio similar to GPT-5.5 in internal evaluations.
.jpg)
