Anthropic's Fable AI Model Pulled from Claude Code Subscriptions Amid Skyrocketing Compute Costs

Anthropic’s highly capable Fable (also known as Mythos) AI model is set to be removed from Claude Code’s Pro and Max subscription plans by June 23rd, effective June 22nd. This decision stems from the model’s significant computational cost and resource intensity, which renders it unsustainable for the heavily subsidized subscription tiers. These plans currently offer users inference capacity valued at up to $8,000 per month for a mere $200-$400 subscription fee. One power user, detailing their experience, reported consuming over $5,470 worth of inference across a laptop and a Mac Mini in just ten days, underscoring the model’s immense utility and the financial scale of the underlying compute. This limited window has spurred developers to creatively maximize their usage before Fable’s departure.

To exploit the remaining subsidized access, developers are employing sophisticated strategies. Techniques include “dual-wielding” multiple Claude Code accounts, automating the triggering of 5-hour session resets via cron jobs and agents like Hermes, and dynamically swapping accounts mid-workflow using the Claude Code CLI. This allows continuous inference across various projects and prevents interruptions when hitting rate limits. Leveraging “Ultra Code” and multi-agent workflows for bulk tasks—such as reviewing numerous GitHub Pull Requests or auditing codebases—demonstrates rapid token consumption, with one workflow burning 1.8 million tokens in under 30 minutes. Furthermore, developers are migrating long-running agent tasks to remote machines like Mac Minis, often accessed via SSH or Tail Scale, to sustain continuous operation without tying up primary workstations. This push is driven by an ambition to redefine developer workflows, allowing single engineers to achieve “10x to 100x” more output by orchestrating complex, multi-agent tasks and exploring unprecedented productivity levels.