Claude Fable 5

Unrivaled Programming Capability: It scored 80.3% on the authoritative SWE-Bench Pro benchmark, far surpassing Opus 4.8 (69.2%) and GPT-5.5 (58.6%). On the more challenging FrontierCode benchmark, it achieved 29.3% —more than double the score of the runner‑up—demonstrating that its code can be genuinely accepted into large‑scale open‑source projects. Stripe used it to migrate a 50‑million‑line Ruby codebase in just one day, a task that would have taken a human team two months.

Powerful Vision and Comprehension: It can replicate the source code of a web application from a screenshot alone, and reaches top‑tier performance in knowledge‑work domains such as financial reasoning and document analysis. Using only pure visual input, it was even able to complete the game Pokémon FireRed.

Genuine "Spatial Intelligence": In one evaluation, Fable 5 was tasked with building a 3D voxel world of the three‑tiered castle of the "Eldian Kingdom" from Attack on Titan from scratch using Three.js. It not only completed the macro‑level framework, but also autonomously applied game‑optimization techniques such as "streaming rendering."

Top-tier developers & engineers – Large-scale codebase migration, complex system refactoring, advanced programming tasks.

Tech companies & startups – Rapid prototyping, automated code generation, faster product iteration.

Finance, legal & research institutions – Complex document analysis, data reasoning, long-horizon research projects.

Game & 3D content creators – Leveraging spatial intelligence for 3D scene construction and game logic design.

Paid users seeking cutting-edge AI – Pro/Max subscribers willing to accept high token costs for top-tier performance.