Anthropic launches Claude Opus 4.8 with focus on honesty, better judgement, and lower AI costs

Anthropic has officially launched its new AI model called Claude Opus 4.8. This is a latest version of its flagship AI model and according to the company, its biggest improvements is not just raw intelligence. Rather, the tech giant is focusing on ‘Honesty’, which the AI users have increasingly been frustrated with. Also Read: ChatGPT 5.5 vs Claude Opus 4.7: Which model should you trust?

As per Anthropic, Claude Opus 4.8 is basically designed to be more reliable while handling uncertain information. The company says, its newly launched model is better at admitting when it does not know something. Interestingly, it will not confidently generate unsupported answers. Also Read: Anthropic’s Claude Opus 4.6 AI model brings bigger context and smarter reasoning

In its official blog post, Anthropic said that early testers noticed a clear shift in the model’s behavior. The company stated that Opus 4.8 is “more likely to flag uncertainties about its work and less likely to make unsupported claims.” Also Read: Anthropic Unveils Claude Opus 4 and Sonnet 4: The Next-Level AI Models for Coding and Reasoning

This is one of the important change at a time when AI hallucinations are rising and remain one of the biggest challenges. The advanced AI chatbot systems are not being honesty in their answers and at some point of time they generate hallucinated responses.

Anthropic believes that its Opus 4.8 is a tangible improvement over its Claude Opus 4.7. As per company’s claims, users noticed a strong and powerful judgments along with more dependable performances during agentic tasks.

These are the tasks where AI systems are expected to independently carry out multiple steps. Additionally, they are involved in decisions and interact with tools and workflows with minimal human guidance.

Anthropic also wants to reduce AI costs

The cost involved in running advanced AI systems has become extremely expensive for companies and as a result businesses deploying AI tools at larger scales are forced to come up with high priced subscription system.

However, Anthropic claims that its new Claude Opus 4.8 model is three times cheaper as compared to the earlier versions o its AI model. The announcement plays an important role at a time when several technology companies like Google, Microsoft, and many others have publicly raised concerns around rising AI infrastructure and operational costs.

In addition, the tech giant has also introduced a new feature called Effort Control, allowing users to choose how much reasoning effort the AI should spend before generating a proper response to user.

To understand this in better way, imagine you selected a high-effort mode if you want a deeper reasoning and more thoughtful answers. In this mode, Claude will spend an additional time in processing information before responding. Besides this, you can also switch to lower mode for faster replies while consuming fewer computing resources.

Dynamic Workflows can coordinate hundreds of AI subagents

The company has also announced a new research preview feature called Dynamic Workflows. The feature is introduced to perform extremely large and complicated tasks. These responsibilities are difficult for a single AI system to manage alone.

Anthropic says Dynamic Workflows enables larger AI models like Claude Opus to run hundreds of AI subagents in parallel. These subagents can collaborate with each other on various components of a large task.

The feature may be useful for software engineering and enterprise level workloads. Anthropic claims that it can support developers to analyze thousands of lines of code, scan for software vulnerabilities in large-scale services or test big file migrations before they go live.

Mythos-class AI capabilities coming soon

Anthropic also provided an update on its next-generation Mythos-class AI features. It’s been a while since the company made headlines in April with its announcement that Mythos could find cybersecurity vulnerabilities much better than previous AI systems.

The announcement had raised concern among the cyber security researchers as there may be some risks involved with the highly functional systems used for discovering vulnerabilities.

Add Techlusive as a Preferred Source

Mythos class features will be available to customers in the next couple of weeks, the company said.