With its latest launches, Anthropic seems to be attempting to make highly capable AI accessible while controlling its potential risks. (Express Image/Anthropic)Frontier AI lab, Anthropic has introduced two new artificial intelligence (AI) models, Claude Fable 5.1 and Claude Mythos 5.1. The company claims that Fable 5.1 is built to improve coding, knowledge work, and research, while Mythos 5.1 is positioned particularly for cybersecurity and biology. While the two models are based on the same underlying system, they have different safety safeguards. Fable 5.1 is generally available, and Mythos 5.1 is offered only to a small set of vetted users and organisations working in domains like cybersecurity and life sciences.According to Anthropic, Fable 5.1 comes with a significant performance improvement over its predecessor, Fable 5, especially in terms of complex, long-running tasks. The company has also focused on making the model more cost-efficient. Anthropic said that typical workloads are expected to cost around 25 per cent less than Fable 5, while highly agentic workloads could be up to 45 per cent cheaper.The lower cost is mainly because Anthropic has cut the price of cache reads, which allow the model to reuse information it has already processed. Cache reads will now cost $0.25 per million tokens, 75 per cent less than before. The price for regular input and output remains unchanged at $10 and $50 per million tokens, respectively.When it comes to performance, Anthropic’s own benchmarks show Fable 5.1 performing strongly across several areas. It secured 52.6 per cent on the Terminal-Bench-Science benchmark, compared with 24.7 per cent for Fable 5 and 29 per cent for Opus 5. On CursorBench, which evaluates agentic coding capabilities, Fable 5.1 scored 73.4 per cent, compared with 70.5 per cent for Fable 5 and 70 per cent for Opus 5.Also Read | OpenAI says upcoming model is so capable it requires stronger guardrailsIn the announcement blog, the company also highlighted examples of the model solving difficult real-world problems. For instance, investment firm Millennium used Fable 5.1 to probe a rare software crash that had remained unexplained for several years. According to Anthropic, the model traced the problem to a bug located in an external vendor library.Another major focus of the new models is scientific research. Anthropic said Mythos 5.1 was able to design high-affinity protein binders in testing. Across 12 targets, almost 50 per cent of its designs were confirmed as viable binders, compared with typical hit rates of around 10-15 per cent in protein design, according to the company.Further, the company said that Fable 5.1 also worked on planetary science. Anthropic said the model trained a neural network to create a higher-resolution elevation map spanning about one-third of Venus, using data from NASA’s Magellan mission. The new map could reveal details at about two to three kilometres, compared with 10 to 20 kilometres in the earlier map.Story continues below this adMeanwhile, Anthropic is also introducing changes to the models’ safety systems. The AI firm said that its updated cybersecurity safeguards produce around 60 per cent fewer interventions per session for Claude Code users compared to the safeguards used with Fable 5. The model is also capable of identifying software vulnerabilities for defensive security work, although certain activities, including exploit generation and penetration testing, remain limited.On the other hand, Mythos 5.1 is available through Anthropic’s trusted access programmes for selected cybersecurity and life sciences professionals. The company said it plans to expand access in the future. Fable 5.1 is available now across Anthropic’s platforms, including Amazon Web Services, Google Cloud and Microsoft Azure.Also Read | AI models are getting better at training other models, Anthropic study findsWith its latest launches, Anthropic seems to be attempting to make highly capable AI accessible while controlling its potential risks. Anthropic’s Fable and Mythos line began with Claude Mythos Preview in April this year, which was made available to a limited group of cyber defenders through its Project Glasswing. The launch also raised concerns about how quickly AI could change the cybersecurity threat landscape. Anthropic found that the model could autonomously detect previously unknown vulnerabilities, including those in major operating systems and browsers, and in some instances turn them into working exploits.A widespread concern was that attackers could likely use similar AI systems to find and exploit vulnerabilities much faster and with less technical know-how. Following this, the company also warned that the gap between detecting a vulnerability and exploiting it could shrink considerably, leaving software developers and users less time to patch flaws. This is said to be the primary reason Anthropic decided against releasing Mythos Preview to the public.Story continues below this adThe company later followed it with Fable 5 and Mythos 5 in June, making Fable the safeguarded version for general use and the less-restricted version of Mythos for trusted partners. The new Fable 5.1 and Mythos 5.1 are the next step in this approach, with improved capabilities and more precise safeguards designed to reduce unnecessary blocks and false positives.