What’s changed, why it matters to your work, and something useful to try next.
Aimed squarely at people doing formal research — scientists or analysts who need to run multi-step literature reviews or lab data analysis and be able to show their working. Claude Science ties together lab tools, databases and notebooks so the output comes with an audit trail, which matters if your results need to hold up to peer review or compliance scrutiny rather than just be a quick summary.
For most day-to-day AI-assisted work — drafting, analysis, coding help — Sonnet 5 is the practical choice right now: it gets close to the top-tier Opus model's performance at a lower cost, which matters if you are choosing a model for regular use rather than occasional heavy-duty tasks. Worth locking in current workflows on it before the introductory pricing rises later in the year.
Relevant to anyone producing marketing, training, or social visuals on a budget, as these are faster, lower-cost models for generating images and short videos with conversational editing. Also relevant to teams building their own tools via the API who want a cheaper option for image or video generation.
Relevant if you use or were waiting on Fable 5 — access is back online now with a stronger safety classifier behind it. The broader jailbreak-severity framework being built with Amazon, Microsoft and Google is more of an industry safety signal than something that changes how you use the model day to day.
Mostly infrastructure housekeeping — if you had older conversations running on GPT-4.5, they've automatically moved to the newer GPT-5.5 model. Nothing for you to do.
Not yet actionable for most people since access is currently limited to trusted partners, but worth knowing a next-generation model family — a flagship plus faster, cheaper variants — is on the way, likely to affect pricing and speed options once it becomes generally available.
If you use ChatGPT to help with personal budgeting or spending questions, that feature is no longer limited to a small pilot group — it now works for anyone on the Plus plan, including on Android. Dictation (talking instead of typing) is also more accurate across more languages.
Useful for anyone studying toward a qualification or upskilling independently: upload your own course materials and get a diagnostic quiz, a personalised lesson plan, and a progress dashboard, rather than working through content unguided.
Anyone hosting a site built with Manus can choose to pay nothing while it is idle (good for a site with occasional traffic) or keep it always-on for busier ones, switching with one click as needs change — a simple way to control hosting costs.
Purpose-built for lawyers and legal teams — it connects directly to the legal research databases, document systems and contract tools they already use, so research and drafting can happen without re-keying information between separate platforms.
If you use ChatGPT to help weigh up a decision, plan something, research options, or think through a purchase, everyday conversations with the fast "Instant" model should feel more helpful and considered — this improvement applies automatically, nothing to switch on.
If your team runs on Slack, Claude Tag lets you bring Claude into a channel the same way you would loop in a colleague — tag it on a thread and it picks up the conversation's context to help with the task, plus it can proactively flag relevant updates without being asked. Useful for reducing the back-and-forth of copying context into a separate AI chat window every time a question comes up in a team channel.
SCALA’s updates draw on published provider announcements. Each article separates the source summary from practical interpretation. Steve’s guides and the SCALA method help you take the next step.
Meet Steve and SCALA