No vendor shipped anything on 1 August, which makes it a good day to read the month as a whole.
Three structural shifts run through July's changelogs. None of them is "the model got smarter" — that happened too, and it mattered least.
One: nobody is betting on a single lab any more
Copilot added GPT-5.6 on 9 July, Gemini 3.6 Flash on 21 July, Claude Opus 5 on 24 July and Grok 4.5 on 28 July. Cursor shipped Router on 22 July to move requests between models by cost and difficulty.
The two leading tools independently reached the same position within a week of each other: model choice is a runtime decision, not a company identity.
Two: the floor fell out of pricing
OpenAI cut GPT-5.6 Luna eighty percent and Terra twenty percent on 30 July. Cursor launched a ₹649 India plan on 28 July.
Both moves reward volume. The kind of feature that runs a model over every row, every message and every document just became a normal thing to build rather than a budget conversation.
Three: models leave, and that is now scheduled
GitHub published August deprecations on 31 July, naming Gemini 2.5 Pro and Gemini 3 Flash. On the same day Enterprise gained team-level model policy targeting, and two days earlier default model enablement changed.
Model policy became an administrative lifecycle in one week: what arrives, who gets it, and what disappears.
What to take from it
If you run a team, three things follow directly.
- Do not couple a workflow to one named model — changing model should be configuration, not a project
- Re-check your usage bill quarterly; July's prices invalidated June's assumptions
- Give model policy an owner, because it now both adds and removes capability from your developers