Methodology

Anthropic recently released the Sonnet 4.6 model, now available through Cloud Code. The presenter set out to compare Sonnet 4.6 with the existing Opus 4.6 model using a consistent experimental framework, motivated by Boris’s claim that Sonnet “nears Opus‑level intelligence.”

Methodology

The test suite comprised seven Laravel projects: five default starter kits (React, Vue, Livewire), one API‑only project, and one Filament admin‑panel project. For each codebase the models were asked to generate CRUD files, refactor the database schema, and fix hidden bugs. Evaluation relied on automated external tests and a manual review of code quality. The experiment ran sequentially—first Opus completed all tasks, then Sonnet tackled the same set. Session usage started at 0 %; after Opus it rose to 37 %, and after Sonnet to 49 %.

Performance Results

Opus required 39 minutes to finish the seven projects, while Sonnet completed them in 26 minutes, a clear speed advantage. Token consumption, however, was higher for Sonnet (49 % of the session) than for Opus (37 %). In terms of test outcomes Sonnet achieved zero failed tests, whereas Opus produced a single failure caused by a non‑existent create_first method in a seeder.

Code Quality Analysis

Opus tended to dive deeper into the codebase, employing more object‑oriented patterns, class‑based validation rules, and the latest Laravel features such as Wayfinder syntax. Sonnet often used older syntax but still produced functional code. On the UI side Sonnet showed strengths: it leveraged the Flux library for richer button and icon components and made more intuitive menu‑item groupings in the admin panel. Opus, by contrast, generated more detailed database queries, for example including post counts when listing categories.

Interpretation and Conclusion

The higher token usage of Opus appears linked to its deeper analysis, frequent consultation of documentation, and adherence to best practices. Sonnet’s faster turnaround likely stems from a leaner approach that skips extensive guideline checks. The presenter questions whether the modest quality differences matter to most project owners, noting community feedback that Sonnet suffices for roughly 95 % of everyday work. For small Laravel projects Sonnet is cheaper and quicker, making it the recommended daily‑work model, while Opus remains valuable for more complex tasks where thoroughness outweighs speed and cost.

Future Plans

The presenter will continue running comparative experiments with upcoming models and encourages readers to subscribe to AI Coding Daily for updates. Earlier comparisons with open‑source alternatives are available to premium subscribers.