Theo Browne compares Opus 5 with Fable 5 and GPT-5.6 Sol using published evaluations and his own development work. Theo Browne separates token price from tokens consumed, elapsed work and subscription limits, arguing that a lower advertised price does not imply a proportional reduction in task cost.
Theo Browne describes competing plans for a T3 Code change, cross-model reviews and subsequent implementation work. The examples show why creator preferences about maintainability, thoroughness and instruction following can differ from benchmark rankings. Model-generated review scores are not independent proof that one plan is best.
Theo Browne explains why Opus 5 currently feels like a useful compromise between diligence and code quality for his projects, while also showing an unwanted browser action and an incorrect explanation from the model. Theo Browne's discussion of safety and retention policies reflects reported release behavior and personal experience, not a guarantee of compliance for every deployment.
Theo Browne concludes by recommending representative side-by-side tasks and external review rather than blindly adopting his preference. The comparison remains specific to his codebases, harnesses and usage patterns, with Fable 5 and GPT-5.6 Sol retaining strengths in other work.
Watch on YouTube
