Anthropic Reportedly Tests Fable 5.5 Claude Model as Google Runs Gemini 4 Carbon Internally

Anthropic is reportedly testing a new frontier Claude model with early users ahead of an expected launch in the coming weeks, according to reporting from The Information cited by 36Kr. While the report did not officially confirm whether the model belongs to the top-tier Fable family or second-tier Opus family, a GitHub bot under Microsoft Foundry and the open-source project LiteLLM both briefly registered the identifier "claude-fable-5-5" on October 13 before deleting the entries. Testers granted early access told The Information that the model surpasses all currently available Anthropic models in reasoning and intuition, particularly in identifying unspoken user objectives from minimal prompts. Instead of merely patching a single reported software bug, the model reportedly traces the underlying defect category across a codebase and asks the user for permission before applying broader fixes. Developers who were reportedly routed to the stronger model during early gray-scale testing noted that it ran for 2 hours straight and consumed 27M tokens to generate a frame-by-frame JavaScript animation from a four-line prompt, and outperformed GPT-6.1 Sol in a 3D jellyfish physics test. While online rumors suggested an immediate Tuesday release, The Information reporter Stephanie Palazzolo clarified that the launch is expected in the coming weeks. Meanwhile, following the release of Gemini 4 Argon, Google has deployed its next model, Gemini 4 Carbon, on its internal programming platform Jetski, according to internal documents and screenshots viewed by Business Insider. OpenAI is also reportedly preparing its next-generation Terra model for launch.