Claude Opus 4.8 launched, Anthropic says honesty and sharp judgment are top features
Anthropic has unveiled Claude Opus 4.8, saying it is better at admitting uncertainty and avoiding unsupported claims. The update also brings cheaper fast mode, Effort Control and Dynamic Workflows as the company pushes larger AI tasks.
Just 41 days after launching its previous AI model, AI startup Anthropic is back with another upgrade, and this time, the company says the biggest improvement is not just intelligence, but honesty. The company has unveiled Claude Opus 4.8, the latest version of its flagship AI model. The model delivers the expected improvements in coding, reasoning, and general knowledge. But Anthropic says the real change addresses something many AI users have been complaining about: AI models making claims that they can’t support.
According to Anthropic, early testers found that Opus 4.8 is more willing to admit uncertainty and less likely to make unsupported claims.
“Early testers report that Opus 4.8 is more likely to flag uncertainties about its work and less likely to make unsupported claims,” Anthropic wrote in the release announcement.
Anthropic described Opus 4.8 as a “modest but tangible improvement” over Opus 4.7, but the company claims it performs better in agentic tasks. Early testers reportedly found the model sharper in judgement and more reliable during such tasks.
Anthropic also wants to reduce AI costs
With Opus 4.8, Anthropic is also trying to tackle another growing problem in the AI world: cost. Opus 4.8’s fast mode is said to be three times cheaper than previous Anthropic model versions. That matters because companies are increasingly worried about the enormous cost of running advanced AI systems. In recent days, firms including Microsoft and Uber have raised concerns about soaring AI spending.
Along with this, Anthropic has also introduced a new feature called Effort Control. The feature allows users to decide how much “thinking power” the AI should use before answering a question.
If users choose a high-effort setting, Claude will spend more time reasoning deeply before responding. If they pick a lower setting, the AI will answer faster while consuming fewer resources and rate limits. In practical terms, it works a bit like switching between “quick reply mode” and “deep thinking mode.”
Dynamic Workflows aims to handle bigger AI tasks
Another major addition is Dynamic Workflows, currently available in research preview. The feature is designed for extremely large and complicated tasks that cannot realistically be handled by a single AI agent.
Anthropic says Dynamic Workflows allows larger AI models like Opus to coordinate hundreds of parallel AI subagents at once. It can be useful for tasks such as checking thousands of lines of code, hunting software bugs across an entire service, or testing a migration touching hundreds of files before release.
Mythos-class capabilities in the coming weeks
Back in April, the company had alarmed cybersecurity experts after revealing that Mythos was dramatically better at discovering vulnerabilities than previous models. Anthropic now says it has made rapid progress on safeguards and expects to bring Mythos-class capabilities to customers in the coming weeks.
Just 41 days after launching its previous AI model, AI startup Anthropic is back with another upgrade, and this time, the company says the biggest improvement is not just intelligence, but honesty. The company has unveiled Claude Opus 4.8, the latest version of its flagship AI model. The model delivers the expected improvements in coding, reasoning, and general knowledge. But Anthropic says the real change addresses something many AI users have been complaining about: AI models making claims that they can’t support.
According to Anthropic, early testers found that Opus 4.8 is more willing to admit uncertainty and less likely to make unsupported claims.
“Early testers report that Opus 4.8 is more likely to flag uncertainties about its work and less likely to make unsupported claims,” Anthropic wrote in the release announcement.
Anthropic described Opus 4.8 as a “modest but tangible improvement” over Opus 4.7, but the company claims it performs better in agentic tasks. Early testers reportedly found the model sharper in judgement and more reliable during such tasks.
Anthropic also wants to reduce AI costs
With Opus 4.8, Anthropic is also trying to tackle another growing problem in the AI world: cost. Opus 4.8’s fast mode is said to be three times cheaper than previous Anthropic model versions. That matters because companies are increasingly worried about the enormous cost of running advanced AI systems. In recent days, firms including Microsoft and Uber have raised concerns about soaring AI spending.
Along with this, Anthropic has also introduced a new feature called Effort Control. The feature allows users to decide how much “thinking power” the AI should use before answering a question.
If users choose a high-effort setting, Claude will spend more time reasoning deeply before responding. If they pick a lower setting, the AI will answer faster while consuming fewer resources and rate limits. In practical terms, it works a bit like switching between “quick reply mode” and “deep thinking mode.”
Dynamic Workflows aims to handle bigger AI tasks
Another major addition is Dynamic Workflows, currently available in research preview. The feature is designed for extremely large and complicated tasks that cannot realistically be handled by a single AI agent.
Anthropic says Dynamic Workflows allows larger AI models like Opus to coordinate hundreds of parallel AI subagents at once. It can be useful for tasks such as checking thousands of lines of code, hunting software bugs across an entire service, or testing a migration touching hundreds of files before release.
Mythos-class capabilities in the coming weeks
Back in April, the company had alarmed cybersecurity experts after revealing that Mythos was dramatically better at discovering vulnerabilities than previous models. Anthropic now says it has made rapid progress on safeguards and expects to bring Mythos-class capabilities to customers in the coming weeks.
