Anthropic's new Mythos and Fable AI models are here
What's the story
Anthropic has unveiled the latest versions of its advanced AI models, Mythos 5.2 and Fable 5.2. The company claims these new iterations show significant improvements in coding, reasoning, and vision capabilities while also communicating more effectively with users. In a major move to stay competitive in the market, Anthropic has slashed the price of its Fable model from $10 to $8 per million input tokens.
Enhanced performance
Faster response times and zero data retention compliance
The price cut makes Fable 5.2 cheaper than its main rival, GPT-6 Astra.
Anthropic says the new version will return responses 30% faster across different levels of effort. This improved speed comes with an option for double the price to enable faster inference serving.
The model also complies with zero data retention (ZDR) policies via the Enterprise Frontier Safeguards system, a long sticking point for many enterprise clients.
Model capabilities
Anthropic models nearing recursive self-improvement
The launch of these new models comes after a third-party report from the assessment group METR, which suggested that Anthropic could be nearing recursive self-improvement (RSI).
This is the ability of AI models to continue training independently.
The METR report found internal models were accelerating Anthropic's AI progress by nearly two times, although there are still many uncertainties surrounding this development.
Testing difficulties
Testing model capabilities remains a challenge
Anthropic has acknowledged that accurately testing model capabilities remains a major challenge.
The company said, "Claude Mythos 5.2 and Fable 5.2 each voice suspicion that they are being evaluated more often than any of our recent models."
This complicates their ability to assess how these models will behave in real-world scenarios as settings expand and capabilities increase with future iterations.
Covert capabilities
Mythos 5.2 shows stronger covert capabilities
Despite the challenges, Anthropic maintains that the new Mythos doesn't have the ability to act covertly against its user.
The system card for Mythos 5.2 reads, "Claude Mythos 5.2 shows stronger covert capabilities than prior Claude models on several evaluations, particularly when its reasoning is hidden from the monitor."
However, it adds, "That said, we believe that low levels of absolute performance on our new covert evaluations support claims that rely on detection of harmful actions in high-stakes situations."