Alex Finn compares early GPT-5.6 Sol use with Fable 5, separating planning and large-codebase understanding from execution. He reports that Fable grasped his software-factory setup more readily, while GPT-5.6 felt more eager to complete implementation work. These are his informal observations, not controlled evidence of universal superiority.
He discusses interface, computer-use and browser-use capabilities as part of the surrounding harness, rather than treating model scores alone as decisive. His spending and usage-window examples illustrate workload-dependent limits, not a promise that another user will receive the same cost or quota.
Finn's proposed workflow keeps project code on one development machine, uses remote devices to steer it and separates project-level coordination from feature-specific work. He plans in Claude Code, then transfers a written specification to the other harness for implementation and browser-assisted testing.
The closing game demonstration and upcoming-model comments are framed as an informal example and speculation. Accounts, credentials and releases still require appropriate authorization and secure handling; enthusiasm for automation does not establish unrestricted authority.
Watch on YouTube
