AI Pricing Changes in Q4 2026: 8 Flagged Repricings, Checked Against the Page

This autumn, automated price trackers reported that Runway raised its top AI video plan by 25% while sales AI vendors Sybill and Regie.ai cut theirs. We opened all three vendors' own pricing pages, then five more flagged AI price changes. Of 8 AI pricing changes in 2026's busiest repricing season, 0 held up exactly as reported. Two of them were the same monthly versus annual billing toggle, read in opposite directions.

The AI pricing changes headlines that did not survive a second look

The topic we set out to write was a clean split by category. AI video was repricing up, led by Runway's top plan moving from $76 to $95. AI sales tools were discounting into competition, with Regie.ai and Sybill cutting their top tiers.

It does not survive contact with the pricing pages. Every one of those three numbers is printed on the vendor's site today. The trouble is that the "before" and "after" prices are both live at once, on the same page, separated by a billing switch.

That matters beyond three vendors. Automated price trackers are now a common input for procurement teams and pricing analysts. If you are using one to decide whether a category has pricing power, you need to know how often its flags are real. Our answer, on this sample, is rarely.

How these AI pricing changes were checked

We started from the public changelog published by CostBench, a tracker that says it re-extracts each vendor's public pricing page and compares it against its last verified snapshot. We picked every AI product among its largest September movers and its October entries where the claim was a price move rather than a review note. That gave 8 vendors across video, sales, research, transcription, presentations and developer tooling.

For each one we opened the tracker's entry, recorded the claimed old price, new price and date, and then opened the vendor's live pricing page on 20 October 2026. Where a second tracker, PricingSaaS, logged the same vendor, we read its record too. We classed each claim into one of four outcomes.

The four outcomes are simple. A billing toggle is a change where both prices are on the page now, one billed monthly and one billed annually. A plan mismatch compares two different plans, or two plans on different billing bases. A real repricing is a list-price or packaging change confirmed by a second record. Unverifiable means the vendor page did not contain the claimed figures at all.

One limit belongs here rather than in a footnote. We tried to open archived copies of each pricing page from the Internet Archive, and the snapshots were not retrievable in this pass. Our verdicts therefore rest on what the live page shows today. For the two toggle cases that is enough, because both figures are live. For the others it is weaker, and the table says so.

Riverside's page carried a staging-environment notice when we read it, so that row is marked low confidence. Runway was logged twice by the same tracker, under two product names, on 7 September and 8 October. It is counted once.

Runway and Sybill: one billing toggle, two opposite price headlines

The two most quoted moves in the set are the same mistake.

Runway's $95 Max plan has always sat beside $76

Runway's current pricing page lists four plans. Standard is $12 a month billed annually, Pro is $28, and Max is $76, each with a struck-through monthly figure of $15, $35 and $95. The page's toggle reads "Yearly -20% off", and the Max card says billing annually saves $228 a year, which is 12 times $19.

The tracker's claimed change is $76 to $95, or +25%. Divide 95 by 76 and you get exactly 1.25. A 20% annual discount, read backwards, is always a 25% increase. Nothing about Runway's price list needed to move for that flag to fire.

The tracker's own price-history page for Runway disagrees with its changelog. It records a range topping out at $95 in March 2026, then $76 from April through its 13 October snapshot, with no September entry. Its changelog says the opposite happened in September. Two pages from one tracker, two directions.

Sybill's Business plan is $90 and $108 at the same time

Sybill's pricing page renders two price sets. One shows Pro at $30 and Business at $90 per user a month. The other shows $36 and $108. The only billing text on the page is a toggle reading "Annual Save 17%".

The flagged change, $108 to $90 on 11 October, is a 16.7% decrease. That is the annual saving on the page, to the decimal. A different tracker, which listed Business at $90 billed annually, never reported a change at all.

We think the right reading is that Sybill's prices did not move. The page renders both sets, and an extractor that captured the monthly state one week and the annual state the next would log exactly this. We could not see the archived page to prove it, which is why the verdict is "billing toggle" and not "Sybill never changed price".

Regie.ai is the one real AI price change in the set

Regie.ai's pricing really did change. It changed shape, not just level, and it happened about six weeks before the date in the flag.

What actually changed on Regie's pricing page

Until late August, Regie sold two publicly priced plans on annual contracts: AI SEP at $180 per user a month and Force Multiplier Rep at $499. PricingSaaS compared captures from 17 and 24 August 2026 and recorded both plans discontinued and replaced by self-serve RegieGO tiers. On 28 September it logged a new Team plan and a Free tier expanded to 10 people sharing one credit pool.

Regie's live page now lists Free at $0, Pro at $41 or $49 a month, Team at $125 or $150 per seat with a two-seat minimum, and custom Enterprise. Pro and Team are self-serve with no contract, and the toggle advertises 17% off for annual billing.

Why "Regie cut its price" is the wrong summary

The tracker logged the top tier falling from $499 to $150 on 7 October, a drop of about 70%. That compares a contracted, seat-minimum sales plan with a credit-metered self-serve plan. They are not the same product at two prices.

The better reading is that Regie moved its front door downmarket. Its entry price for a paid seat fell from $180 on an annual contract to $49 with no contract. Its heavy users now buy credits, at 15,000 a seat per month on Team, rather than a fixed seat with a fixed scope. That is a change in what you pay for, and it is covered from the buyer's side in the breakdown of why AI credits confuse buyers.

Credit-metered plans tend to look cheaper per seat and cost more per heavy user. Our view is that you should model your top 3 users' monthly credit burn before your renewal, not after.

Five more flagged AI price changes, five more problems

The remaining five flags are less famous and fail in more varied ways.

Elicit was logged as raising its top tier from $49 to $169. Its pricing page still sells Pro at $49 per user a month billed annually. The $169 figure belongs to Scale, a separate collaboration plan. A plan was removed from the lineup, which is a real event, but no plan went from $49 to $169.

Riverside was logged as a 191% rise from $34 to $99. On the page we read, $34 is the Grow plan billed annually and $99 is the Webinar plan billed monthly. That compares two plans on two billing bases. The page also carried a staging notice, so we hold this one loosely.

Simplified was logged as a 229% rise from $24 to $79. The $79 is its entry Starter plan billed annually, and the top plan is Business at $199 billed annually. The headline confuses the bottom of the price list with the top.

Speechmatics was logged at $0.24 to $0.38 per minute for a plan called Flow. The page has no Flow plan and no $0.38 figure. It does show $0.24 as an hourly rate for one model, which suggests a unit error. We cannot confirm what the tracker saw.

Weights & Biases was logged as a 140% rise from $25 to $60 per user. Its pricing address now redirects to a CoreWeave page, so the claim stays unverified.

None of this is a criticism of a single tracker. The methodology page of the tracker we used lists these exact failure modes itself: promotional rates read as list, annual compared against monthly, renamed or restructured plans, regional variation and per-seat versus per-account billing. It says that when it hand-checked 16 candidate changes, 12 were artefacts and only 2 were real. Our sample of 8 landed in the same place.

Where AI pricing is genuinely moving up in Q4 2026

If page diffs cannot tell you which categories have pricing power, announcements can. The increases we could verify this quarter came with a press page, an effective date and a reason.

Atlassian's 13 October increase is the clean example

Atlassian publishes its upcoming cloud list prices in advance. Its tables show Confluence Cloud Premium for 1 to 100 users rising from $13.20 to $14.10 per user per month, effective 13 October 2026, with existing customers paying the new price on renewals from that date. Larger tiers move too: 251 to 1,000 users goes from $7.60 to $8.15.

The stated reason is broader than AI. Atlassian's FAQ says it is updating prices "to reflect the latest innovations" and describes AI as built into every paid Jira edition, alongside Loom, Rovo and Bitbucket changes. We think the honest description is a platform increase with AI as the lead justification, not an AI surcharge.

A separate tracker run by Gravity, which only counts increases where the vendor itself names AI, logged 45 AI-linked list-price increases from 9 vendors since September 2024, with a median rise of 15.4%. Its method is the opposite of a page diff. It starts from the vendor's announcement and only then records a price. It also states its largest gap: vendors who raise prices by customer email alone are missing.

The pattern is the one we would bet on. Incumbents with installed bases raise list prices on renewal and call it AI. AI-native challengers, like Regie, restructure into self-serve credit plans to win new logos. If you want the renewal side of that story, it is covered in how grandfathered pricing turns into a renewal increase.

Where this repricing tracker is weakest

This post makes a strong claim from a small sample. Here is everything that should make you discount it.

The sample is small and was chosen from the noisiest end

We checked 8 vendors, chosen from one tracker's largest September movers and its October AI entries. Large percentage moves are exactly where artefacts concentrate, because a monthly versus annual mix-up or a plan mismatch produces bigger swings than a real 5% list increase. A random sample would almost certainly show a higher share of real changes. The 0 of 8 figure describes the headline flags, not the whole market.

We did not see the archived pages

The method we set out to use was to check each claim against an archived copy of the vendor's page from before the flagged date. The archive snapshots exist for Runway, Sybill and Regie, and we could not open them in this pass. Our toggle verdicts hold without them, since both prices are live today. Our plan-mismatch verdicts are inferences from the current page, and a vendor could have changed a plan and then changed it back. The Weights & Biases and Speechmatics claims could be entirely accurate.

There is also a fair case for the tracker. Its changelog flags candidates at volume, across thousands of products, and its own methodology warns readers about every failure we found. The fault is less in the tool and more in quoting its entries as findings.

A quarterly AI repricing tracker that survives its own audit

We still want the quarterly tracker the topic promised. It just has to be built from rules that would have caught all 7 failures above. These are the ones we will use from Q1 2027, and you can use them for your own vendor list today.

Rule 1: test the ratio against the toggle. Before logging any change, divide new by old and compare it with the vendor's advertised annual discount. If a 20% discount sits on the page and the change is +25%, or a 17% saving sits there and the change is −16.7%, it is a toggle until proved otherwise. This single test would have caught both Runway and Sybill.

Rule 2: compare a named plan, never the top of the range. "Top tier" changes whenever a vendor adds, drops or renames a plan. Elicit, Simplified and Regie were all logged as top-tier moves that were really lineup changes. Track each plan by name, and log a rename as a rename.

Rule 3: require two captures or one announcement. A change counts only if it appears in two captures at least a week apart, or in the vendor's own announcement.

Rule 4: record what the unit is. Per seat, per account, per credit, per hour and per minute are different products. Speechmatics shows what happens when the unit slips. Our earlier look at how AI vendors cut usage limits without touching price shows the reverse case, where the price holds and the unit shrinks.

Rule 5: publish the misses. Every quarter, the tracker should list the flags it rejected and why. A repricing dataset that only shows confirmed changes looks cleaner than it is. The rejections are where readers learn which numbers circulating elsewhere are wrong.

Our broader read of what repricing teaches is in the lessons from 1,800 SaaS pricing changes, and a per-vendor history for one AI product is laid out in the ChatGPT plan pricing timeline.

Frequently asked questions

Where to start with your own AI vendor prices

Open your three largest AI subscriptions' pricing pages today and write down four things for each plan you use: plan name, monthly price, annual price and billing unit. Save the page as a PDF with today's date. That file is the "before" snapshot you will need the next time a vendor, a tracker or a colleague tells you the price moved.

Then check whether any of those vendors has a published future-pricing page or renewal notice, as Atlassian does. An announced increase with an effective date is the only kind you can budget for. A flagged change on a tracker is a question to ask your account manager, not an answer.

Related reading