Tech • IA • Crypto
Code with Claude Tokyo 2026: Opening Keynote
9.4/10TL;DR
Anthropic unveiled its most advanced AI models, Claude Fable 5 and Claude Mythos 5, alongside new agent-building tools aimed at accelerating autonomous software development and enterprise workflows.
KEY POINTS
Launch of Claude Fable 5 and Mythos 5
Anthropic introduced Claude Fable 5 as its most capable generally available model, with Mythos 5 offered under restricted access due to higher-risk capabilities. Both models represent the fifth generation of Claude systems and significantly expand performance across coding, research, and multimodal tasks. The release reflects rapid progress in AI capability, with shorter development cycles between major leaps.
Breakthroughs in coding and autonomy
Fable 5 demonstrates strong gains in single-shot correctness, solving complex engineering problems in one attempt, and long-horizon autonomy, maintaining coherence across tasks that can run for days. Early testing shows it can complete work equivalent to weeks of human effort from a single prompt, while also orchestrating sub-agents efficiently and cost-effectively.
Advanced code understanding and debugging
Beyond generating code, the model excels at analyzing existing systems. It can trace outages, navigate large repositories, and identify improvements proactively. Earlier iterations of the technology have already uncovered long-standing vulnerabilities, including a 27-year-old flaw in OpenBSD, highlighting its potential in software security.
Expansion beyond coding into enterprise workflows
Fable 5 is designed for real-world business operations, handling documents, financial analysis, and multi-step processes end-to-end. It performs particularly well on messy, ambiguous inputs, enabling it to manage complex organizational tasks and produce professional-grade outputs without tight constraints.
Enhanced vision and multimodal capabilities
The model shows improved accuracy in interpreting technical diagrams, charts, and web interfaces, extending its usefulness in engineering, scientific, and design contexts. This positions it as a broader reasoning system rather than a narrowly focused coding assistant.
Safety controls and restricted capabilities
Due to risks in areas like cybersecurity and biology, Anthropic introduced a routing system that redirects sensitive queries to safer models. Mythos 5, which removes these safeguards, is limited to vetted researchers under Project Glasswing, balancing openness with risk mitigation.
Rapid growth of AI-driven development platforms
Anthropic reported a 17× year-over-year increase in API usage, signaling strong developer adoption. Companies such as Rakuten and Canva are using Claude to accelerate product cycles, with some teams shifting from quarterly releases to biweekly deployments and enabling non-programmers to generate functional applications.
Introduction of managed AI agents
New Claude-managed agents provide developers with infrastructure, memory, and execution environments to build autonomous systems. Features include scheduled deployments, secure credential handling, and iterative goal-based execution, allowing AI systems to act rather than just suggest.
Self-improving agent capabilities
Agents can store memory, develop reusable “skills,” and perform reflective optimization through a feature described as “dreaming,” where systems review past runs to improve future performance. This supports the emergence of persistent, self-improving workflows.
AI-native enterprise vision
The platform promotes a model where work is executed directly by AI systems, with humans defining goals rather than performing tasks. Examples include autonomous debugging, automated reporting, and continuous optimization processes running without manual intervention.
Developer productivity gains
Anthropic’s tools are reshaping software development workflows. Internal engineers reportedly produce eight times more code, while companies like Spotify and Mercari have used Claude to automate large-scale migrations and boost engineering output by up to 90%.
Parallelized workflows and large-scale automation
New features such as dynamic workflows allow hundreds of agents to operate in parallel on tasks like localization, refactoring, and audits. Complex operations that once required hours of manual coordination can now be executed from a single prompt.
CONCLUSION
Anthropic’s latest models and platform updates signal a shift toward fully autonomous, AI-driven software and business operations, with developers increasingly acting as orchestrators of intelligent systems rather than direct implementers of code.