AI Tool Changes in Q2 2026: What Changed in Your Stack and What to Do Next

Sora shut down. Google retired Whisk. ChatGPT added another Pro option. Some Perplexity Pro users saw lower advanced model limits. A field guide to the AI tool changes that actually hit the tools you're paying for, with the Choosely read on what to do about each.

Quick take: Nobody emails you every time the AI tool you depend on gets pricier, gets weaker, or disappears. In Q2 2026, all three showed up in the tools people actually use. The fix is not following AI news harder. It is treating your stack as a set of decisions that quietly expire, then running a simple AI stack audit before those changes catch you. Below: what changed, who it affects, and what to do about it.

You chose your AI tools carefully. You compared the options, picked the ones that fit, and moved on. The choice was not the mistake. The "moved on" was.

Because the tools did not move on. They kept changing underneath you, and almost none of those changes arrived with a notification. The price crept up. The free tier thinned out. A feature you had built a workflow around got deprecated. Occasionally the whole product just ended.

Q2 2026 has been a blunt reminder of exactly this. We pay attention to these shifts at Choosely because they are the heart of what we do: recommend the right tool for a job clearly, and keep that recommendation honest as the ground moves. Treat this less as "AI news" and more as a field guide to the changes that hit the tools you already pay for, each with a plain English read on what to do next.

What changed at a glance

• Sora was discontinued. This hits video creators, marketers, and API users. Export anything still available and rebuild the workflow.
• Whisk moved into Flow. This affects Google Labs users and image or video experimenters. Download or migrate assets, and avoid making Labs tools critical to your work.
• ChatGPT pricing and model access shifted. This matters for power users, Codex users, and teams. Check your actual plan, model access, and limits.
• Some Perplexity Pro users reported lower advanced model limits. This mostly affects researchers and heavy daily users. Check whether your plan still fits your real usage.
• Consolidation kept rolling through AI tools. This matters for anyone relying on mid tier specialist apps. Keep your exits clean and your alternatives known.

A flagship got switched off: OpenAI shut down Sora

OpenAI announced the end of Sora in late March, but the user impact landed in Q2. The web and mobile apps went dark on April 26. The API follows on September 24, after which OpenAI says account data associated with Sora usage will be deleted.

This lands hardest on creators and marketers who built any part of a video workflow on Sora, and on anyone who assumed a famous, well funded tool was a safe long term bet. What it breaks: your library, if you do not export what is still available in time, and any pipeline that called the API. Reporting from TechCrunch and the Wall Street Journal points to a less dramatic but very instructive reason: usage fell, compute costs stayed heavy, and OpenAI redirected attention toward higher priority work. Famous and funded did not make it permanent.

Choosely's read: Export anything you can still access immediately, then rebuild on a tool matched to the job, not the brand name. Veo for realism, Kling for speed, and Runway for control are the usual landing spots. We walked through a real multi tool video stack in Chloe vs History, and you can compare the current video options in the Choosely tool browser. Then bank the lesson: vendor size is not vendor permanence.

Google quietly retired Whisk

Less dramatic, same lesson. Google confirmed that Whisk, its Google Labs text to image blending tool, moved into Flow and was discontinued on April 30. Google warned that any remaining media in a user's Whisk library after that date would be permanently deleted and no longer recoverable. Anything labeled Labs, preview, or experiment can be pulled at any time. That is fine for trials and risky as a foundation.

Choosely's read: If a "Labs" tool has quietly become critical in your work, that is a risk flag, not a feature. Keep a non experimental backup for anything you actually depend on.

ChatGPT rewrote its pricing again

The fast version: GPT-5.5 launched on April 23, and GPT-5.5 Instant started rolling out in May as ChatGPT's new default model. OpenAI also added a $100 per month Pro option, sitting between Plus and the highest usage $200 Pro plan. Ads are now part of the Free and Go picture in selected markets, starting with testing in the US and expanding gradually. Codex access has also moved from temporary promotional allowances toward the standard 5x and 20x Pro usage structure. In other words: "what ChatGPT plan am I on?" is no longer a complete answer.

The implications depend entirely on who you are:

• Casual users: "free" can now mean ads, and the model behind your chats can change silently. If you mostly use ChatGPT for everyday jobs, a specialized tool is often cheaper and better than any tier. Here is where specialized beats general.
• Power users: the $100 Pro option is the real decision. The honest question is whether you will use the extra capability or just buy out of fear of missing out.
• Teams: tier to model mapping is shifting fast, so the model your seats actually run can lag the consumer flagship. Verify what you are on; do not assume.

Choosely's read: Check what your plan includes today versus the day you subscribed: model, limits, perks, ads, and add on usage. "I'm on Plus" tells you a lot less than it used to.

The quiet downgrade: Perplexity Pro felt smaller for some users

This is the sneakiest category of all, because there is no announcement and no new invoice. The thing you already pay for just starts feeling like it does less.

In mid May, some Perplexity Pro users reported their advanced model limits had been sharply cut, with prompts nudging them toward the $2,000 per year Max tier. Perplexity attributed the affected accounts to enforcement against promo code fraud and resale rather than a blanket reduction. Fair enough, but to an affected user, the plan they bought still felt smaller. The people this hits hardest are researchers and heavy daily users, exactly the ones who paid for Pro to get the higher ceilings.

Choosely's read: Quiet downgrades do not ping you, so trust the feeling when a tool starts seeming stingier or pushier about upgrades. The "best research tool" is a question worth asking again, not answering once. Here is our current take.

The pattern underneath: consolidation

Zoom out and the dominant move of the last year plus is not shutdowns. It is absorption. Bigger platforms keep buying capabilities they cannot build fast enough, and mid tier AI tools keep getting rolled into larger suites. An acquisition rarely kills a tool overnight. It does something quieter. The roadmap stalls, pricing gets redirected toward the acquirer's customers, and your scrappy favorite slowly becomes a feature inside a product you did not choose.

Choosely's read: Assume any mid tier tool you adopt today might not exist as a standalone in 18 months, and do not let a single tool become a dependency you cannot exit cleanly.

How to keep your stack honest

Put the four together: shutdowns, price moves, quiet downgrades, and consolidation. The conclusion is simple: the best tool for a job is a moving target, and the decision you made about it starts expiring the moment you make it. The answer is not to churn tools every month. Churn has its own cost. The answer is to review your stack on a cadence instead of assuming it is settled.

Run these three questions on your stack this week:

1. 1For every AI tool you pay for, what model and limits does my plan actually include today, versus when I signed up?
2. 2For every tool holding my work, could I export everything in ten minutes if it shut down tomorrow?
3. 3For my top three categories, if I ran the decision again today, would I still pick what I am using?

If the answer to that last one is "not sure," that is the exact gap Choosely is built to close. We recommend the right tool for a specific job, clearly, with the tradeoffs in plain English, rather than handing you a generic answer you have to interpret yourself.

And Stack Intelligence is now live in beta: it remembers the tools you have saved and flags material changes worth your attention, including shutdowns, pricing shifts, quiet downgrades, and major capability changes, so a quarter like this one does not slip past you. It is deliberately high signal. The goal is to catch changes that genuinely affect your stack, not push you into constant tool churn.

Founding member accounts are free right now, no card required, and they lock in 50% off Pro for life when paid plans launch. Start a free account and build a stack Choosely can help you keep honest.

Q2 2026 made the point loudly: in AI, choosing once is not choosing. It is just the first decision. Keeping your stack honest is the rest of the job.

Sources: OpenAI on the Sora discontinuation; Google on Whisk moving to Flow; OpenAI on GPT-5.5; OpenAI Pro tier guidance; OpenAI Ads in ChatGPT FAQ; TechCrunch on why OpenAI shut down Sora; Wall Street Journal reporting summarized by Investing.com; and Android Authority on Perplexity Pro limits. Figures and dates are accurate as of June 2026 and, fittingly, are subject to change.