July gave us GPT 5.6, Claude Opus 5, Sonnet 5, and the return of Fable 5.
Every new release reshuffles the benchmarks - but I m beginning to think the model is only half of the experience.
The harness around it increasingly decides how productive you actually are:
How well it understands your codebase
How it manages context across longer tasks
Whether you can run and review multiple agents
How clearly it presents diffs and changes
How easily you can switch models or continue work elsewhere
Whether it fits into your existing subscriptions and workflow