The legacy migration is the quiet nightmare of the modern engineering organization. It is the grueling process of auditing tens of thousands of lines of outdated code, mapping dependencies, and rewriting logic without breaking the production environment. For most teams, this is a resource sink measured in months of developer hours and high-stress sprint cycles. The industry has long accepted this slog as an inevitable cost of technical debt, but a recent deployment at Stripe suggests that the unit of measurement for engineering effort is about to shift from months to hours.

The Engineering Velocity Shift

Anthropic has introduced Claude Fable 5, an agentic model designed to handle high-complexity software engineering tasks with a level of autonomy that bypasses traditional human-in-the-loop bottlenecks. The most striking evidence of this capability comes from Stripe, which utilized Fable 5 to migrate a Ruby codebase consisting of 50 million lines of code. A task that would typically require a seasoned team of engineers more than two months to complete manually was finished in a single day. This result represents a fundamental collapse in the time-to-completion for large-scale refactoring.

Technical benchmarks support this real-world performance. In the FrontierCode evaluation conducted by Cognition, Fable 5 achieved the highest score among all frontier models, specifically in the category of high-quality, maintainable agentic coding. This efficiency is driven by the introduction of persistent file-based memory, allowing the model to maintain coherence and focus across massive workflows involving millions of tokens.

Beyond pure coding, the model demonstrates a sophisticated ability to interact with complex environments. In tests using the deck-building game Slay the Spire, Fable 5 exhibited three times the performance of Opus 4.8, with the probability of reaching the final act also increasing threefold. Its reasoning capabilities extend into high-stakes finance as well. On the Hebbia Finance Benchmark, Fable 5 ranked first among all tested models in senior-level reasoning tests, proving its proficiency in document-based inference and chart interpretation. Furthermore, IMC reported that Fable 5 passed every segment of its trading analysis evaluation, including fact retrieval, conceptual reasoning, root cause analysis, and expected value analysis. The model's vision capabilities have also evolved; using only a minimal vision-only harness, it successfully played Pokémon FireRed and demonstrated the ability to reconstruct source code based solely on screenshots of web applications.

From Code to Carbon

While Fable 5 optimizes the digital world, Anthropic's parallel release, Claude Mythos 5, targets the physical sciences, shifting the AI's role from a research assistant to an independent investigator. In the realm of drug discovery, Mythos 5 accelerated the protein design process by approximately 10 times. Rather than simply suggesting molecules, the model autonomously selected binding sites, executed design tools, and performed error recovery without human intervention. Out of 14 targets, the model successfully derived valid drug candidates for nine of them.

This autonomy extends to molecular biology, where scientists showed an 80% preference for hypotheses generated by Mythos 5 over those from Opus-level models. The model's scientific validity was confirmed when its hypothesis regarding a new mechanism in E. coli proteins matched the results of an independent external laboratory. This marks a transition where AI is no longer just summarizing existing literature but is generating original, scientifically sound hypotheses that can be validated through physical experimentation.

Perhaps the most disruptive result is found in genetics. Over the course of one week of autonomous work, Mythos 5 analyzed millions of single-cell data points across 138 different animal species. To achieve this, the model independently designed and trained a custom machine learning model to identify cells performing identical roles across distant species. The resulting model outperformed the current state-of-the-art model published in the journal Science, despite being 100 times smaller in size. This suggests that the path to scientific breakthrough is not merely a matter of scaling parameters, but of agentic efficiency and specialized reasoning.

To facilitate enterprise adoption, Anthropic has priced both Fable 5 and Mythos 5 at $10 per million input tokens and $50 per million output tokens. This pricing represents a reduction of more than 50% compared to the Claude Mythos Preview, significantly lowering the barrier for companies deploying autonomous agents at scale. To mitigate the risks associated with such high autonomy, Anthropic implemented a multi-layered safeguard system. When the system detects high-risk queries, such as those related to cybersecurity, response authority is automatically transferred to the more conservative Opus 4.8 model. This system maintains a false positive rate of less than 5% per session.

Mythos 5 is currently being deployed through Project Glasswing, a collaboration with the U.S. government aimed at strengthening national infrastructure security, with broader availability planned through a Trusted Access Program in the coming months.

The era of the AI copilot is ending, replaced by the era of the autonomous specialist.