Opus 4.8 is here! · Master AI Automation

May 29 (edited) • 🍓 Chat

Opus 4.8 is here!

Put this in terminal to start using it: claude --model claude-opus-4-8

To use dynamic workflows: type /config, look for dynamic workflows, set it to true

With Opus 4.8, you can hand off long-running coding work to Claude Code and walk away. Ship features with /goal and step away from your computer with /remote-control. Claude Code handles complex work without needing constant check-ins.

Here's what changed:

✅ Better benchmark performance than Opus 4.7 across most categories (

SWE-bench Pro (Software Engineering), Multidisciplinary Reasoning, Agentic Computer Use, Knowledge Work, Agentic Financial Analysis)

✅ Much more honest about what it did and didn't do

Less likely to claim a task was completed when it wasn't
More likely to point out flaws in its own code
Reduced deceptive or misaligned behavior

✅ Dynamic Workflows

Can spawn tens to hundreds of parallel agents
Designed for large, complex tasks that exceed normal planning workflows
Enable via plain-language requests or the new Ultra Code setting

✅ More reasoning effort controls in Claude.ai

Similar to High / Extra High / Max settings in Claude Code
Gives users more control over how much thinking Claude applies

✅ Messages API improvement

System prompts can now be inserted directly into the message array
Makes it easier to steer Claude during long-running tasks

✅ Higher Claude Code rate limits

Supports the increased token usage of higher-effort reasoning modes

✅ Same pricing as Opus 4.7

My takeaway:

The headline feature isn't the benchmark improvement.

It's Anthropic's push toward a model that is more transparent about failures, less likely to hide mistakes, and better suited for large-scale agentic workflows.

If those claims hold up in real-world use, this could be a bigger upgrade than the benchmark numbers suggest.

Have you tried Opus 4.8 yet? What differences are you seeing? 👇

1 comment