Anthropic unveils Claude Opus 4.8; teases Claude Mythos rollout in weeks

Anthropic has announced an upgraded version of its artificial intelligence (AI) model called Claude Opus 4.8. The model is built on Claude Opus 4.7, which claims to offer improved performance across the board, including coding, agentic skills, reasoning, and practical knowledge. The company touts the new AI model as a “more effective collaborator,” with improvements across benchmarks.

Alongside the release of Claude Opus 4.8, Anthropic also teased a broader rollout of its cybersecurity model, Claude Mythos. It is currently accessible to over 50 companies under the “Project Glasswing,” but now it may make a public release in the coming weeks, according to Anthropic’s blog post.

Must read: AI spending nightmare: Companies spend over a $500 million in 30 days on Anthropic’s Claude

Claude Opus 4.8: Performance and upgrades

The Claude Opus 4.8 is said to be a more efficient AI collaborator that not only works with you but also makes better decisions along the way. Based on its early testers, the company claims the model is more capable and reliable in performing agentic tasks, and that it is more cautious and self-aware. 

The model is said to flag uncertainties and is “less likely to make unsupported claims.” Anthropic said, “Opus 4.8 is around four times less likely than its predecessor to allow flaws in code it has written to pass unremarked.”

In terms of performance, the Claude Opus 4.8 scored 69.2% in Agentic coding (SWE-Bench Pro), up from Opus 4.7’s 64.3% score. In Multidisciplinary reasoning, it scored 57.9%, up from 54.7%, offering improved ability to solve complex, cross-domain problems using external tools. In Agentic computer use (OSWorld-verified), the model achieved 83.4% in performance, up from 82.8%. 

Must read: OpenAI no longer the most valuable AI startup as Anthropic hits $965 billion

When it comes to research and knowledge work, the model performance increased from 1,753 to 1,890. In addition, the company claims that Claude Opus 4.8 Fast Mode is 2.5× faster, making it faster in task completion. 

Claude Opus 4.8 features

Alongside performance gains, the Claude Opus 4.8 also brings new features such as dynamic workflows in Claude Code that claim to manage complex programming tasks more effectively than before. It can breakdown big coding job into many smaller parts and run multiple AI “subagents” at the same time.

The model also brings effort control in Claude.ai and Cowork that lets users adjust effort settings. “On higher effort settings, Claude will think more frequently and more deeply to give better responses. On lower effort settings, Claude will respond faster and use up a user’s rate limits more slowly,” the post said. Lastly, Claude lets developers make changes to instructions mid-task, from permissions to token limits, without stopping the process.