Anthropic’s A highly productive AI model it’s in the hands of a few select organizations and cyber security experts, but it’s the most powerful Claude model you can use now to get the upgrade.
Claude Opus 4.8, released Thursday, is a “modest but noticeable improvement” over Opus 4.7, Anthropic said in a blog post.
But the company also said it is making significant progress in producing a version of its Claude Mythos Preview model that it is ready to release to the public. Currently, it has limited access to Mythos to a consortium of partners as part of what it calls Project Glasswing, explaining that the capabilities of the cybersecurity model have advanced enough to warrant giving cybersecurity experts and large technology companies some initial time to patch the flaws found by the model.
“Models of this level of capability require strong cyber protection before they can be widely released,” Anthropic said. “We are making rapid progress in developing these defenses and expect to be able to deliver Mythos class models to all of our customers in the coming weeks.”
Myths for everyone?
Anthropic’s decision to withhold Mythos previews from the general public, at least for now, was an interesting decision. Was it a smart, forward-thinking move to protect critical Internet infrastructure from potential failures? Was it an easy way to renew the marketing hype? Security researchers have found that the model is certainly able to detect exploits much faster than human hackers, even if it doesn’t push beyond human capabilities. The latest version of Mozilla’s Firefox includes more than 200 fixes identified by Mythos Preview.
Read more: AI Arms Race Accelerating with New Models from OpenAI, DeepSeek and Anthropic
But the fact that the Mythos will soon be available to anyone, even with important cybersecurity guardrails, means that we will finally see that the model lives up to the hype, with all the risks that may involve.
Darren Williams, founder and CEO of cybersecurity firm BlackFog, told CNET via email that major model releases are often tense moments.
“On the other hand, Anthropic’s decision to plan the release, to hold back until defenses are developed, shows the right direction,” he said. “But the more competent the model, the greater the stakes if those defenses fail or if the model is misused. The window between the release of a powerful model and the widespread adoption of defenses against it is always a vulnerable time.”
Mythos will be more expensive to use than other AI models, however, and that could limit its use for hackers. Jake Williams, a cybersecurity researcher and faculty member at IANS Research, said the Mythos was 30 times more expensive in testing than the previous Opus model.
“This is beyond the reach of many, including any horror actors,” Williams told CNET via email. “Country actors already have better technology to detect threats. This only changes the game for a small percentage of threat actors.”
What’s new with Claude Opus 4.8
As for Opus 4.8, Anthropic said it is an improvement in all benchmarks compared to Opus 4.7. The test found that Opus 4.8 is less likely to make unsupported claims and may exhibit uncertainty, the company said.
Several new features are also coming to Anthropic’s AI products, including the ability to control how much “effort” a model will use to respond to input from Claude.ai and Claude Cowork. A higher effort will probably get you better results as the model spends more time responding, but it will heat up to your limits quickly. A lower setting will respond faster and reach level limits more slowly.