From July through September I logged 33 changes, and 22 landed in September alone. API prices mostly fell while subscription limits mostly tightened. Below are the patterns, the changes worth checking first, and a free sheet to match against your own stack.
If you pay for AI tools, your bill shifted this quarter, whether or not you spotted it.
From July 1 to September 30, 2026, I recorded 33 pricing and usage-limit changes across the AI vendors I follow, and 22 of them hit in September. Some were straightforward price cuts. Some were limit cuts dressed up as increases. A few were so quiet you'd only notice when a workflow died halfway through a task.
I covered several as they happened, including Three vendors restructuring usage limits in two weeks. This post gathers the full quarter in one place, along with a free sheet you can filter down to the tools you really use.
The whole quarter, summed up
API prices fell. Subscription limits tightened.
The sheet breaks down like this:
9 changes lowered costs, nearly all of them API or model prices
5 raised prices, mostly GPU compute and creative tool plans
5 cut access, mostly subscription limits and free plans
4 restructured how usage gets counted, which tends to make costs harder to compare
The remainder were new model launches, temporary boosts, mixed changes, and one retirement
Builders on APIs mostly got good news in Q3. Anyone on a flat-rate subscription got a reminder that promotional limits don't last.
Cheap AI has never been cheaper. Generous AI plans have never been harder to predict.
The changes to check first
Of the 33, these are the ones I'd look at first if you use the tools involved.
Claude Code: a +25% that's actually a 17% cut
Anthropic ran a +50% weekly limit promotion on Claude Code across the summer and extended it on July 19. On September 14, a permanent +25% baseline took its place. Digital Applied's limit ledger points out that this is a real gain over the original limits, yet a 17% reduction from what people had during the promo. Both descriptions are accurate, and which one you feel depends on when you signed up.
GitHub Copilot: promotional credits expire
On September 1, Copilot Business and Enterprise lost their promotional included credits. Monthly credits fell from 3,000 to 1,900 **on Business and from **7,000 to 3,900 on Enterprise, a cut of 37% to 44%. If your team leans on premium models in Copilot, open your usage reports now.
ChatGPT's five-hour limit returns
On July 12, OpenAI temporarily lifted the rolling five-hour limit for Plus, Business, and Pro users. Its return was announced for July 30, then pushed back. It finally came back on August 26, and the $100 and $200 Pro plans were exempt.
OpenAI's steep API cuts
On July 30, GPT-5.6 Luna fell 80% to $0.20 input and $1.20 output per million tokens. GPT-5.6 Terra fell 20% to $2 and $12, according to CloudZero's OpenAI pricing guide. For high-volume jobs such as classification or extraction, Luna at that price shifts the math on plenty of projects. Bulk parsing of scraped pages, the kind of work I do, is exactly what it suits.
Claude Opus 5.5 arrives cheaper than Opus 5
Claude Opus 5.5 launched at $4 input and $20 output per million tokens, under Opus 5's $5 and $25. Google's Vertex AI listed it on September 23, and Cursor swapped Opus 5 for Opus 5.5 at the lower rate on September 28. In late August Anthropic also cancelled a planned Sonnet 5 price increase, so it stays at $2 and $10.
Free plans are shrinking
Two quieter moves deserve a mention.** Replit removed its free Starter plan** on September 11, which makes the $20 Core plan the new entry point. And Pika's free plan now includes zero monthly credits. For some creative and coding tools, the "free forever" era is closing.
GPU compute got pricier
Renting GPUs now costs more. RunPod raised Secure Cloud rates, with the H100 SXM at $3.49 an hour. Nebius announced an H100 increase from $3.85 to $4.50 an hour, effective October 1. Fireworks AI lifted its US-only serverless premium from 10% to 50%.
Other changes worth a glance
These missed the top list, but one of them could be the one that hits your stack:
Apify cut its Starter plan from $29 to $19 a month on September 11. Good news if you run scrapers there, as I do.
Klaviyo swapped credit tiers for a $20 monthly plan with $10 top-ups on its AI Composer.
Suno limited commercial-use rights to your monthly download allotment.
Moonshot AI retired Kimi K2.5 and the entire moonshot-v1 API family on August 31. Anything pointed at those endpoints is now broken.
Beautiful.ai raised Pro from $12 to $14.50 a month on annual billing, and Hedra raised Basic from $15 to $20.
Zhipu AI added a Team tier to its GLM Coding Plan and set Individual Pro at $80 a month.
OpenAI launched GPT-6 Astra at $10 input and $50 output per million tokens, and Google launched Gemini 3.8 Flash at $0.75 and $3.75.
Each is small alone. Taken together, they show how often the ground shifts under anyone building with AI.
Three patterns behind the numbers
1. Promotions settle below their peak
Digital Applied's ledger found a steady pattern across 2026: temporary boosts get extended a few times, then land below the promoted level but above the original baseline. Claude Code and GitHub Copilot both followed it this quarter. Expect more of the same.
2. "Increased limits" often means "we changed how we count"
On August 24, Cursor told subscribers it had raised included usage but never published the size of the increase, and it moved Auto billing from a flat rate to per-model rates. Z.ai switched its coding plan from prompt counts to token credits. Atlassian re-metered automation as "steps." None of this is automatically bad. Still, once the unit changes, comparing old costs to new ones gets difficult, and that usually works in the vendor's favor.
3. Competition is driving API prices down quickly
Look at the evidence: Luna's 80% cut, Opus 5.5 priced under Opus 5, Perplexity's Fast Search API at 80% below its standard rate, cheaper embeddings from Voyage, and a 69% cut on Together AI's large-model fine-tuning. If you haven't re-checked your API model choices since spring, you're likely overpaying.
Announced and enforced are two different things. Check the date a change takes effect, not the date it was posted.
How to use the free sheet
The Sheet lists all 33 changes with the date, vendor, product, type of change, what changed, the direction for you, and a source link. Here's how to get value from it in ten minutes:
Filter by vendor down to the tools you pay for.
Enter "Y" in the "Affects me?" column for anything touching your stack.
Note one action per row: switch models, change plan, or ignore.
Open the Summary tab, which counts changes by direction and month, plus how many you flagged as affecting you.
Confirm on the vendor's page before acting, because prices move fast.
What I'd do before Q4 budgets
If you manage AI spend for yourself or a team, here's a short action list:
Re-price your API workloads. Run your top three tasks on the newly cheaper models. Luna, Terra, and Opus 5.5 all changed the options.
Check your coding tool limits. Claude Code, Copilot, and Cursor all changed this quarter. Look at real usage rather than what the plan page promised in June.
Leave disappearing free plans before they cause trouble, especially if a client workflow depends on one.
Lock in GPU capacity if you rent compute and your provider's prices are climbing.
Schedule a 15-minute monthly review of a pricing tracker such as UsagePricing's activity log.
A note on sources
The sheet draws on public trackers, chiefly Digital Applied's coding-plan limit ledger, UsagePricing's activity log, and CloudZero's OpenAI pricing guide, along with vendor coverage. It covers the changes that touch the tools most builders, freelancers, and small teams use, which means some vendors are missing. Each row links to its source, so you can verify anything that matters to you.
If I missed a change that hit your stack, tell me in the comments and I'll add it to the next quarterly sheet.
Ten minutes before Q4
Q3 2026 produced 33 AI pricing and usage-limit changes worth knowing. The short version: API prices fell, subscription limits tightened, and several "increases" were really cuts from a promotional peak.
You don't have to follow all of it. You just need to know which rows touch your stack. Download the free sheet, filter it to your tools, and give it ten minutes before you plan next quarter's budget.
Enjoyed this read? Follow me for more stories like it. You can also reach me on LinkedIn or Telegram @kawsarlog for updates on my work.
I take on freelance scraping projects, so if you need data collected, message me on LinkedIn or email hi@kawsarlog.com
Top comments (0)