GPT‑5.6

GPT-5.6 (Claude Fable 5) demonstrates state-of-the-art performance on nearly all tested benchmarks of AI capability, showing exceptional results in software engineering, knowledge work, vision, and scientific research.

The model excels in software engineering tasks, capable of compressing months of engineering work into days. It can perform codebase-wide migrations in large codebases and scores highest among frontier models on difficult coding tasks requiring high-quality production code.

GPT-5.6 (Claude Fable 5) is the new state-of-the-art for vision tasks. It can extract precise numbers from scientific figures, rebuild web app source code from screenshots, and complete complex vision-only tasks like playing Pokémon FireRed with minimal scaffolding.

The model maintains focus across millions of tokens in long-running tasks and improves its outputs by using its own notes. Its performance with persistent file-based memory is significantly better than previous models, enabling it to handle complex, multi-day projects.

GPT-5.6 (Claude Fable 5.1) is designed for long-running agents and multi-day autonomous projects. It supports multiagent orchestration, allows agents to iterate against pre-defined exit criteria, and uses a memory tool for progressive improvement across sessions.

A variant of GPT-5.6, Claude Mythos 5, offers the strongest cybersecurity capabilities globally. It is deployed with lifted safeguards for cyberdefenders and infrastructure providers, initially through Project Glasswing in collaboration with the US government.