Anthropic has unveiled Claude Opus 4.8, a significant upgrade to its predecessor, Claude Opus 4.7, boasting enhanced capabilities for coding, agent work, reasoning, and knowledge work. The platform is accessible via claude.ai, Claude Code, and the Claude API, with the API name claude-opus-4-8.
One of the key features of Claude Opus 4.8 is the ability for users to control the level of effort applied to a response, directly impacting the number of tokens the model will utilize. Additionally, Claude Code now features dynamic workflows, which enable the planning of work, execution of parallel sub-agents, verification of outputs, and reporting back to the user. The Messages API also accepts live changes to the messages array, allowing developers to update instructions during a task without disrupting prompt cache use or requiring a separate user turn.
The pricing structure for Claude Opus 4.8 remains unchanged, with costs of $5 per million input and $25 per million output tokens when not in ‘fast’ mode. Fast mode is available for $10 per million input tokens and $50 per million output tokens, offering a 2.5x speed increase. Anthropic has positioned Opus 4.8 as an ideal solution for coding and agentic workflows, where the model can leverage tools within a context and validate its own work.
Opus 4.8 has undergone testing by several companies across various industries, including software development, law, finance, and research, with overwhelmingly positive feedback on its agentic workflows and cost parity with other models. The platform has also demonstrated lower rates of deception and is less likely to pass flawed code without commentary.
The effort control feature enables users to manage the trade-off between quality, speed, and token burn rates. While Opus 4.8 defaults to high effort, users can opt for ‘xhigh’ for tasks that demand more computation. To support the resulting higher token usage, Anthropic has increased Claude Code rate limits.
Claude Code’s dynamic workflows are designed to cater to large codebases and can seamlessly migrate codebases comprising hundreds of thousands of lines. These features are currently available in research preview and can be accessed on the Enterprise, Team, and Max plans.
Photo by Jan van der Wolf on Pexels
Photos provided by Pexels