30 Days of AI App Pricing: What Actually Changed

Every week we re-verify the prices behind our 50 app reviews against vendors' own pages, and every change lands in our changelog with a source. That habit produced a dataset nobody was aiming for: a documented, month-long record of how AI app pricing actually moves. From early August to early September 2026 we logged twelve pricing-relevant changes — and they don't move the way you'd guess. Headline prices mostly fell. Your bill probably didn't.

The ledger

• Replit quietly cut Core from $25 to $20/month ($17 annual at the current promo) — no announcement post; we caught it on a routine pricing fetch.
• Suno turned on download caps September 3: 7 lifetime on free, 20/month on Pro, 60 on Premier, retroactive to your whole library — and the new terms tie commercial rights to paid-plan downloads (perpetual once downloaded). Extra downloads are sold separately.
• Claude Code replaced its expiring +50% weekly-limit promo with a permanent +25% over the old baseline — which Anthropic itself concedes is a 17% reduction versus what users had, effective September 14.
• OpenAI cut GPT-5.6 Sol API pricing over 20% — $5→$4 input, $30→$20 output — but only through November 21, and consumer subscriptions were explicitly excluded. (GPT-5.6 Luna had already been cut 80%.)
• ChatGPT Business shipped Premium seats August 25 at $125/user/month ($100 annual) — marketed as 'the $100 seat.'
• ChatGPT Pro quietly became two tiers ($100 and $200/month) with the GPT-6 Astra launch, with weekly GPT-6 Pro caps of 50 and 200 messages respectively.
• Google priced Gemini 3.7 and 3.8 Flash at an intro $0.75/$3.75 per million tokens that doubles January 1, 2027 — the discount has an expiry date printed on it.
• Anthropic launched Fable 5.1 at the same headline price as Fable 5 but cut cache reads 75% — an estimated 25% saving on typical workloads, up to ~45% on agentic ones, without touching the sticker.
• Anthropic also cancelled Sonnet 5's scheduled increase: the $2/$10 'introductory' rate became the standard rate, and the planned September 1 hike to $3/$15 never happened.
• Superwhisper made all local Whisper models free on macOS and expanded its trial from 15 minutes to 3,000 words — a genuine price cut to $0 for on-device dictation.
• Microsoft put its 365 Copilot Business add-on on promo — from $21 to $18/user/month — for the first year, through December 31, for orgs under 300 seats.
• OpenAI's GPT-6 Astra debuted in the API at $10/$50 per million tokens — 2.5× GPT-5.6 Sol's discounted rate, setting the new frontier ceiling.

Pattern one: headline prices fell or froze

Count the direction of the sticker moves: two API price cuts, one subscription cut, one promo discount, one cancelled increase, one price held flat with a big cache discount underneath, one product going free. In thirty days, not a single major AI app raised a headline subscription price. The competitive logic is obvious — with three frontier labs shipping monthly, the sticker is the billboard — and it's exactly what a buyer reading only pricing pages would conclude: AI is getting cheaper.

Pattern two: the meters tightened underneath

Now count the effective moves: Suno capped downloads retroactively and tied commercial rights to them. Claude Code's 'permanent 25% increase' nets out to 17% less capacity than users actually had. ChatGPT's newest model arrives with weekly message caps even on the $200 tier — and 'Pro' quietly split into two tiers on the way. This is the month's real story: the subscription price is becoming marketing, and the meter is becoming the price. Caps, credits, allowances, and effort-based billing all moved in the restrictive direction while stickers moved down or nowhere.

Pattern three: discounts now ship with expiry dates

Four of the twelve moves are explicitly temporary: Sol's API cut ends November 21, Microsoft's promo ends December 31, Replit's $17 rate is labeled a promo, and Google printed January 1, 2027 on Gemini Flash's price — the day it doubles. Intro pricing used to be a startup trick; it's now standard practice at the biggest vendors, which means anyone budgeting an AI stack should diarize the cliffs, not just record the prices.

What a buyer should do with this

• Price the meter, not the tier. Before adopting anything agentic or generative, find the cap — downloads, credits, weekly messages, effort billing — because that's where your real bill lives. Every review on this site documents it under 'the catch.'
• Diarize the expiry dates. November 21, December 31, January 1. If your budget assumes today's rate past those dates, it's wrong on schedule.
• Download what you own. Suno's retroactive caps are the warning shot: on platforms moving to metered exports, work you haven't exported is work you don't fully control yet.
• Check prices against the index, not memory. One of this month's twelve changes was announced nowhere at all — we found it because we re-fetch pricing pages weekly under The Receipts Standard.

We'll run this ledger again next month — the receipts accumulate on their own. If you track a price we missed, tell us; corrections make the dataset better for everyone.