Anthropic Unveils Claude Fable 5.2 and Claude Mythos 5.2 AI Models
Anthropic has officially introduced its latest advancements in artificial intelligence, featuring the Claude Fable 5.2 and Claude Mythos 5.2 models.
These new iterations boast substantial enhancements in coding capabilities, logical reasoning, visual comprehension, and scientific problem-solving.
Notably, they operate with increased efficiency, offering faster performance at a reduced cost compared to their earlier versions.
According to the company, both Fable 5.2 and Mythos 5.2 are engineered to undertake intricate technical challenges.
Anthropic reported that during preliminary testing, these models proved instrumental in supporting engineers with software migrations, facilitating the design of electronic devices, advancing robotics assessments, and addressing sophisticated mathematical and scientific inquiries.
Both Claude Fable and Claude Mythos share the same foundational architecture. However, Fable incorporates superior cybersecurity and biological safeguards, rendering it widely accessible.
In contrast, access to Mythos is restricted to organizations engaged in sanctioned cybersecurity and biomedical research, through Anthropic’s verification protocols.
Enhanced Speed and Competitive Pricing
The newly launched models are said to produce results up to 30% faster than their predecessors, Claude Fable 5.1 and Mythos 5.1.
Priced at $8 per million input tokens and $40 per million output tokens, Anthropic also offers an optional fast mode that provides up to 2.5 times quicker token generation, albeit at steeper pricing tiers.
Benchmark Achievements
Anthropic asserts that the latest models surpass their predecessor, Claude Opus 5.5, across a variety of benchmarks that span agentic coding, scientific inquiry, business processes, cybersecurity measures, robotics, visual chart analysis, and multidisciplinary reasoning.
On the FrontierCode 2.0 benchmark, Fable 5.2 recorded a score of 49.1%, which the company claims is the highest among all tested released models.
Furthermore, the company highlighted notable efficiency strides, indicating that Fable 5.2 is capable of executing demanding tasks with fewer tokens than Fable 5.1.
In several benchmarks, it exhibited performance that rivals Opus 5.5 while incurring approximately half the cost per task.
Advancements in Cybersecurity
Anthropic designates Mythos 5.2 as its most proficient model for cybersecurity engineering. In its internal CVE Rediscovery evaluation, it identified 10.2% more vulnerabilities compared to Opus 5.5 and 14.4% more than Mythos 5.1.
Mozilla noted that it has significantly enhanced vulnerability-discovery metrics in comparison to Mythos 5.1.
Fable 5.2’s classifiers are designed to block a majority of high-risk cyber and research biology tasks, reportedly with 55% fewer blocks than Fable 5.1.
Commitment to Safety
Anthropic assesses the likelihood of unsafe behavior occurring in high-stakes contexts as low. Prior to their release, the models underwent rigorous evaluation by the UK AI Security Institute, Frontier Design, and METR.
The company mentioned that the models exhibit a heightened level of suspicion regarding evaluations, a factor that complicates the predictive reliability of testing outcomes in real-world scenarios.
Fable 5.2 is equipped with its most robust protections against illicit data extraction, reducing the information gathered by data distillers by an impressive 90% to 99%.
Additionally, the company has curtailed progress in research capabilities that could expedite AI development.
Availability
Claude Fable 5.2 is currently accessible across all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. Developers can also utilize it under the identifier claude-fable-5-2.
Meanwhile, cybersecurity experts interested in Mythos 5.2 may apply through the Cyber Verification Program, with a separate application process for life sciences researchers.
Enterprise clients can obtain Mythos 5.2 via Claude Security. As an introductory promotion, eligible paid subscribers will receive additional usage of Fable this week.
Source link: Timesofindia.indiatimes.com.
