Last looked in February?
You are five releases behind.
Since early February 2026: Opus 4.6, Sonnet 4.6, Opus 4.7, Opus 4.8, Fable 5 and Mythos 5, then Sonnet 5 and Opus 5. That is not a criticism, it is the shipping pace. This is the catch-up read: what actually changed, and which of it affects work you already do.
Each release brought real gains: longer context, better reasoning, steadily improving tool use. The problem is quieter. Teams that trained on one model’s quirks in February find by July that the control for how hard Claude thinks has changed twice, the pricing has moved, and a workflow that needed careful prompting now works out of the box, while another changed behaviour because a default moved underneath it.
The thinking toggle is gone; effort is the dial
The biggest shift across the period: from manually toggling extended thinking to an effort parameter on a ladder from low to max. It went generally available with Opus 4.6 in February, gained xhigh in April, and is now the primary control on Opus 5, which has thinking on by default. If your training still teaches a thinking on/off switch, that material is out of date. The full walkthrough is in our effort parameter guide.
A million tokens is the new normal
A one-million-token context window, once a Sonnet 4 beta, is now standard across the current range. For document-heavy teams this changes what “read the whole thing” means, and it changes how prompts and retrieval should be structured. What it does and does not change in practice is covered in the context window guide.
Pricing moved, and not in the direction people assume
Opus 5 launched at the same $5/$25 per million tokens as Opus 4.8. Sonnet 5 launched at an introductory $2/$10 through 31 August, with a new tokenizer that counts about 30% more tokens for the same text, so the sticker price and the effective price are not the same thing. Budgets set last quarter need re-checking, not rolling forward.
A whole new model family appeared, then wobbled
Fable 5 and Mythos 5 launched on 9 June as a new family: one generally available, one restricted to approved organisations. Access to both was briefly restricted within the first month and restored on 1 July. New model families deserve a probation period; our adoption guide covers how to run one.
You do not need to chase every release
You need a deliberate, periodic check-in, ideally with someone whose job is tracking these changes so your team does not have to. The releases that matter to you are the ones that touch defaults your workflows rely on: effort, context, tokenizer, pricing windows. Everything else is news, not work.
What is now default behaviour versus what needs configuring has changed more in six months than in the year before it. A one-hour session mapping the changes onto your actual workflows is usually enough to know whether you need a quick refresh or a proper re-training.
Things people
usually ask.
The effort parameter reaching general availability, because it replaced the extended thinking toggle as the way you control reasoning depth, cost and latency on every current model. Everything else is situational; that one touches everyone.
No. But two clocks are ticking: Sonnet 5’s introductory pricing ends 31 August, and older-model conveniences keep being retired, the way fast mode was removed for Opus 4.7 the day Opus 5 launched. Plan the move; do not be forced into it.
A monthly check against a short list of things you actually depend on: your default model, its effort setting, its context window, its pricing, and any beta feature in your workflows. If none of those changed, you are current, whatever else shipped.