Anthropic upgrades Claude AI with Opus 4.8: What's new in flagship model

Claude Opus 4.8 arrives with benchmark improvements, enhanced AI agent capabilities, and new features designed for developers, enterprises, and everyday users

Anthropic has launched Claude Opus 4.8, an upgraded version of its flagship AI model. Building on Claude Opus 4.7, the new release brings better coding, reasoning, agentic capabilities, and knowledge work performance. The company said that the new model is more reliable and a stronger collaborator, while its Fast Mode is now around 2.5 times faster than before. Alongside Opus 4.8, Anthropic has also introduced new features such as Dynamic Workflows in Claude Code, Effort Controls for users, and Messages API enhancements for developers. According to Anthropic, Opus 4.8 is available at the same pricing as its predecessor. Here is what Opus 4.8 offers:

 

Claude Opus 4.8 focuses on stronger collaboration

According to Anthropic, Claude Opus 4.8 delivers improvements across benchmark evaluations covering coding, reasoning, agentic skills, and knowledge-based tasks. The company positions the model as a more effective collaborator that can handle complex workflows while providing better judgment during task execution. 

The company said that early testers found the model to be more reliable when performing agentic tasks. Anthropic noted that Opus 4.8 is better at identifying uncertainty and is less likely to make unsupported claims or overstate progress when evidence is limited. As per the company’s internal evaluations, the model is around four times less likely than Opus 4.7 to leave flaws in generated code unacknowledged.

Improved alignment and safety measures

Before release, Anthropic said it had conducted a detailed alignment assessment of the model. The company said Opus 4.8 achieved stronger results in areas such as supporting user autonomy and acting in the user’s best interests. It also demonstrated lower rates of misaligned behaviour, including deception and cooperation with misuse, compared with Opus 4.7. According to Anthropic, the model’s alignment performance is similar to that of Claude Mythos Preview, one of its most safety-focused AI systems.

 

  READ: CERT-In issues critical severity warning for Google Chrome on desktop users

How Claude Opus 4.8 performs compared to Opus 4.7

• Agentic coding (SWE-Bench Pro): The model’s score increased from 64.3 per cent to 69.2 per cent, indicating stronger performance in software development tasks and autonomous coding workflows.
• Multidisciplinary reasoning (Humanity’s Last Exam): Scores improved from 54.7 per cent to 57.9 per cent, showing better ability to solve complex problems that require reasoning across multiple domains while using external tools.
• Agentic computer use (OSWorld-verified): Performance rose from 82.8 per cent to 83.4 per cent, reflecting incremental gains in carrying out computer-based tasks autonomously.
• Knowledge work (GDPval-AA): The score climbed from 1,753 to 1,890, suggesting improved capabilities in research, analysis, summarisation, and other professional knowledge-based tasks.
• Agentic financial analysis (Finance Agent v2): Results improved from 51.5 per cent to 53.9 per cent, highlighting better performance in financial reasoning and analytical workflows.
• Faster responses: Anthropic said that the Opus 4.8 Fast Mode is approximately 2.5 times faster, allowing users to complete tasks more quickly.
• Lower costs: The company also claims the upgraded Fast Mode is three times cheaper than previous versions, making high-speed AI usage more cost-effective.

Other upgrades

Apart from Claude Opus 4.8, Anthropic has also announced several new features focused on improving coding workflows, user controls, and developer tools. These include: 

Dynamic Workflows: Claude can now handle much larger coding projects by creating and managing hundreds of AI subagents at once, checking their work, and combining the results before responding. It can even help migrate massive codebases with hundreds of thousands of lines of code. 

Effort Control: Users can choose how much effort Claude puts into a response. Higher effort settings deliver deeper reasoning and more detailed answers, while lower settings prioritise speed and conserve usage limits. 

Messages API updates: Developers can now change Claude’s instructions while a task is running, making it easier to adjust permissions, token limits, or context without interrupting the workflow.

  READ: Anthropic valued at $965 bn after latest funding round, eclipsing OpenAI

What’s next?

Anthropic describes Opus 4.8 as a meaningful upgrade over its predecessor but said that it is continuing work on more powerful and cost-efficient AI models. The company also revealed plans to introduce a new class of models with intelligence beyond Opus. 

As part of its Project Glasswing initiative, select organisations are already testing Claude Mythos Preview for cybersecurity applications. Anthropic mentioned that it is developing stronger safeguards before making Mythos-class models more widely available.

Claude Opus 4.8: Availability and pricing

 

Claude Opus 4.8 is available globally starting today. Pricing remains unchanged from Opus 4.7, with regular usage costing $5 per million input tokens and $25 per million output tokens. Fast Mode is priced at $10 per million input tokens and $50 per million output tokens. Developers can access the model through the Claude API.