You approved an AI chatbot or an automation earlier this year, and now there's a charge on the card every month. Then the news broke that OpenAI and Anthropic cut their model prices on September 22, 2026, and one question follows it straight to your desk: should my bill go down too? An AI model price cut can lower what you pay, but usually only one line of your bill moves, and the size of that line depends on how much text your automation sends and receives. The rest of the invoice, the workflow tool, the connectors, the hosting and the support, may not change at all. That gap between the headline and your statement is where the real decision sits. Before you call a vendor or ask anyone to rebuild something that works, you can find out which part of your bill is actually the model. It takes about ten minutes, and you'll know exactly what to ask for.
Key Takeaways
On September 22, 2026, OpenAI launched GPT-6 Sol at $2 input and $10 output per million tokens, and Anthropic launched Claude Opus 5.5 at $4 input and $20 output. ChatGPT and Claude subscriptions are a separate product.
At 1,000 input and 300 output tokens per lead reply, GPT-6 Sol comes to $5 a month in model cost, against $10 at the GPT-5.6 Sol rate.
Make's Core plan is $12 a month and Zapier Professional starts at $19.99, so a model that is 50% cheaper does not make the whole automation 50% cheaper.
Request the model name, tier, monthly token totals, any pass-through markup and the date of the change, then decide.
What did OpenAI and Anthropic change on September 22, 2026?
Both companies released new models on the same day, and both priced them below the models they replace. OpenAI announced GPT-6 Sol and GPT-6 Luna and said both API models cost 50% less than the GPT-5.6 promotional price. Anthropic announced Claude Opus 5.5 with lower input, output and cache rates than Claude Opus 5.
Every price below is in US dollars per 1 million tokens, taken from OpenAI's API pricing page at the Standard tier and from Anthropic's Opus 5.5 page. A token is a small piece of text, often part of a word. "Input" is what your automation sends to the model, and "output" is what the model writes back.
| Model | Input | Output | Replaces | Old input | Old output |
|---|---|---|---|---|---|
| GPT-6 Sol (OpenAI, Standard) | $2.00 | $10.00 | GPT-5.6 Sol | $4.00 | $20.00 |
| GPT-6 Luna (OpenAI, Standard) | $0.10 | $0.50 | GPT-5.6 Luna | $0.20 | $1.00 |
| Claude Opus 5.5 (Anthropic) | $4.00 | $20.00 | Claude Opus 5 | $5.00 | $25.00 |
| Claude Opus 5.5 cache read | $0.20 | n/a | Opus 5 cache read | $0.50 | n/a |
| Claude Opus 5.5 cache write | $5.00 | n/a | Opus 5 cache write | $6.25 | n/a |
A few details on the price pages change the math for real invoices. OpenAI sells the same model at more than one service tier: its Batch and Flex prices are half of Standard, and its Fast tier is double. It also lists separate prices for very long requests. Anthropic prices input, output, cache reads and cache writes separately, and it offers a Fast mode as well. So, the tier your automation runs on matters as much as the model name.
On the Anthropic side, the list-price cut works out to 20% on input and output. Anthropic also says typical workloads on Opus 5.5 cost about 40% less, and it credits that figure to two things together: the lower prices and fewer tokens used per task. That 40% is Anthropic's own claim about its own model. Your workload may use more or fewer tokens, so treat it as a reason to measure, not a number to put on a spreadsheet.
These are API prices, the rates businesses and developers pay when software calls the model directly. They are not a general price cut on every ChatGPT or Claude subscription. If your "AI bill" is a seat on a chat app, this news may not touch it at all.
What do 1,000 AI lead replies cost at the new prices?
The easiest way to see what a per-token price means is to run one month of a common job through it. Take a lead-reply automation: a form comes in, the automation sends the lead's message plus some background about your business to the model, and the model drafts a reply. Say it handles 1,000 leads a month. Assume each request sends 1,000 input tokens and gets back 300 output tokens. That volume is an illustration to show the arithmetic, not a measured average for any business.
Over a month, that is 1,000,000 input tokens and 300,000 output tokens. The formula is the same for every model: divide each token count by one million, multiply by that model's price, and add the two together.
At GPT-6 Sol Standard pricing, that's (1,000,000 ÷ 1,000,000 × $2) + (300,000 ÷ 1,000,000 × $10), which is $2 plus $3, or $5 for the month. The same work at the GPT-5.6 Sol rate was $4 plus $6, or $10.
| Model and tier | Input cost | Output cost | Monthly model cost |
|---|---|---|---|
| GPT-5.6 Sol, Standard (previous rate) | $4.00 | $6.00 | $10.00 |
| GPT-6 Sol, Standard | $2.00 | $3.00 | $5.00 |
| GPT-6 Sol, Batch (half of Standard) | $1.00 | $1.50 | $2.50 |
| GPT-5.6 Luna, Standard (previous rate) | $0.20 | $0.30 | $0.50 |
| GPT-6 Luna, Standard | $0.10 | $0.15 | $0.25 |
| Claude Opus 5, standard input and output | $5.00 | $7.50 | $12.50 |
| Claude Opus 5.5, standard input and output | $4.00 | $6.00 | $10.00 |
The halving is real, but look at what it's halving. On this workload, moving from GPT-5.6 Sol to GPT-6 Sol takes $5 off a month, and moving from GPT-5.6 Luna to GPT-6 Luna takes off a quarter. Those figures come straight from the published prices and the assumed volume; they will scale up or down with your own token counts.
Other jobs follow the same arithmetic with different inputs. Email drafting looks a lot like lead replies. Document question-and-answer tools can be very different, because the automation pulls passages from your documents and sends them along with every question, so input tokens can run far higher than in the example. The only way to know your number is to read the input, cached input and output token totals your automation actually recorded.
Where does the model sit on your AI automation bill?
The model is usually one line on a small business automation bill, sitting next to the tools that run the workflow. An automation also needs something to catch the form, pass the data along, write the reply into your inbox or CRM and log what happened. That something is often a workflow platform, and it has its own meter.
Make's pricing page lists a Core plan at $12 a month for 10,000 credits and a Pro plan at $21. Make's credits documentation says most module actions use one credit each, including reads and writes, so the number of steps in each run decides how fast the credits go. Zapier's pricing starts its Professional plan at $19.99 a month, and each successful step uses tasks. Zapier's rates page says its AI steps use 1, 3 or 5 tasks per step depending on the tier, and if you bring your own model account, you still use Zapier tasks and you still pay the model provider separately.
Put the worked example next to those subscriptions and the bill looks like this:
| Line on the bill | Source of the price | Moves with a model price cut? |
|---|---|---|
| Model use: $5 on GPT-6 Sol ($10 on GPT-5.6 Sol) | OpenAI API pricing, worked example above | Yes, if billed by actual usage |
| Workflow platform: Make Core $12 or Pro $21 | Make pricing page | No |
| Workflow platform: Zapier Professional from $19.99 | Zapier pricing page | No |
| Hosting, database or vector storage | Your provider's plan | No |
| Messaging, monitoring and maintenance | Your vendor or your own time | No |
| Management or support fee | Your contract | Only if the contract says so |
Read down the right-hand column and the answer to "should my bill drop?" becomes specific. In this example, a $5 model saving sits beside a platform plan of $12 to $21 or more, plus whatever hosting and support cost. The $5 Sol figure does not mean a $5 automation. It means a $5 model line.
That doesn't make model cost small everywhere. Long documents, high volume, premium models, tool calls and long outputs can make the model line much larger. The point is to compare your actual line items, not to assume in either direction which one is larger. If you want the wider view of what goes into these systems, our breakdown of how much AI automations cost covers the build and running costs that sit around the model.
Should your vendor lower your price after the OpenAI API price cut?
Sometimes, and the invoice usually tells you which case you are in. A price cut reaches you automatically only when you pay the provider directly for usage. When a vendor sits in between, it depends on how they bill you and what your agreement says. The three parts below cover when to ask, what to ask for, and what to write down so the next change is clearer.
When the cut should reach you
It is reasonable to ask for a lower charge when your invoice shows a separate model or API usage line, when your vendor bills actual provider usage plus a stated margin, or when your proposal named GPT-5.6 Sol, GPT-5.6 Luna or Claude Opus 5 as the model. In those cases, a lower provider rate has somewhere to land.
A fixed monthly managed-service fee is different. It may cover engineering, support, monitoring and platform charges, so a provider cut does not reduce it on its own. You can still ask what the fee includes and whether model use is bundled into it. Also check whether your automation has moved to a new model at all. A system still calling GPT-5.6 Sol is billed at GPT-5.6 Sol rates until someone changes it.
What to ask your vendor for
Ask for the facts that let you check the math yourself:
- Model identifier: The exact model name the automation calls today.
- Service tier: Standard, Batch, Flex or Fast, since each is priced differently.
- Monthly token totals: Input, cached input and output tokens for the last full month.
- Usage record: The provider invoice or a usage export for your account.
- Pass-through markup: Whether model fees are passed through at cost, marked up, capped or bundled.
- Effective date: When any model or price change took effect, or will.
OpenAI's organization usage reporting can group results by model and includes request counts and input and output token totals, so a vendor running your automation on OpenAI has a way to produce this.
What to put in the contract
If you're signing or renewing an AI automation agreement, a few clauses make the next price change easy to read. None of these is a legal requirement; they are commercial terms that make the bill checkable. List platform and support fees separately from model fees. State whether third-party model fees are pass-through, capped or bundled. Require written notice before the vendor changes the model or the pricing tier. Ask for a monthly usage report. And set an approval threshold for overages or for swapping to a materially different model.
No regulator publishes a rule that requires an agency to pass along an AI model price cut. Whether your bill changes comes down to your agreement, which is why these terms are worth settling now.
How can you check your own AI bill in ten minutes?
You don't need anyone's permission to start. Pull up your most recent automation invoice, any OpenAI or Anthropic invoice, and your workflow platform's billing page, then write down:
- Who owns each account: Is the OpenAI, Anthropic, Make or Zapier account in your business's name or your vendor's?
- Which model and tier: The model name, and Standard, Batch or another tier if shown.
- How much it used: Input, cached and output tokens, or the number of model requests, for the last full month.
- What the workflow used: Zapier tasks, Make credits, or the number of executions.
- Any fixed fee: A management, support or hosting charge that stays the same each month.
Where to look depends on the tool. In OpenAI, the organization usage reports cover model, requests and token counts, though you may need the account administrator to open them. In Zapier, check your task usage and whether pay-per-task billing is turned on. In Make, check your credits and the scenario's execution history. Don't hand API keys or customer data to an outside reviewer just to get this report; the billing pages are enough.
Once you have those numbers, run the formula from the lead-reply example with your own token totals and the current price for your exact model and tier. Compare the result with the previous rate. That difference is the most a model price cut can take off your model line, and you can now see it next to everything else you pay for.
If you can't find a model name, a token count or a usage line at all, that is a finding too. A single unexplained "AI fee" with no usage access gives you no way to know whether this week's news affects you. Asking for that breakdown is a fair first request, and it matters more than the discount.
Is it worth switching models or rebuilding now?
Usually not on a headline alone. If the model is a small, bundled or already-updated part of a workflow that runs well, a rebuild can cost more and add more risk than the price cut saves. Look at your own numbers from the bill check first; if the model line is a few dollars, a rebuild is hard to justify on price.
Switching to a newer model is more than changing a label. A new model can use more or fewer tokens for the same task, which is part of why Anthropic's 40% figure depends on token use as well as price. Anthropic has also noted that Sonnet 5's tokenizer can turn the same input into roughly 1.0 to 1.35 times as many tokens, depending on the content. Before moving an automation to a cheaper or newer model, test it on a batch of real inputs and check the output format, any tool calls, the safety rules and what happens when something fails. Then compare the whole bill, not just the per-token rate.
The cheapest model isn't automatically the right one either. GPT-6 Luna costs a twentieth of GPT-6 Sol per token, but whether it writes lead replies you'd be happy to send is something only testing can show. How the provider handles your data is a separate question from price. OpenAI says business and API data is not used to train its models by default, and its data controls page says abuse-monitoring logs may be kept for up to 30 days by default.
There's a useful standard behind all of this. NIST's voluntary AI RMF 1.0, published January 26, 2023, says organizations should have policies and procedures in place and should plan ongoing monitoring and periodic review. Its core functions are Govern, Map, Measure and Manage. It isn't a pricing rule, but it points to the same habits: know which model you run, watch its usage, and review the arrangement when something changes. A price change like this week's is a good time for that review.
Does the September 2026 price cut reach your bill?
Pick an answer to begin.
1. GPT-6 Sol costs 50% less per token than GPT-5.6 Sol. What happens to a Make or Zapier subscription that runs the automation?
2. Your automation handles 1,000 lead replies a month at 1,000 input and 300 output tokens each. What is the model cost on GPT-6 Sol Standard?
3. Which request helps you most before asking a vendor for a lower price?
Frequently Asked Questions About ai model price cut
Did OpenAI lower its API prices in September 2026?
Yes. GPT-6 Sol and GPT-6 Luna launched on September 22, 2026, at 50% below the GPT-5.6 promotional API price. Sol is $2 input and $10 output per million tokens at the Standard tier, and Luna is $0.10 and $0.50.
What is Claude Opus 5.5 pricing?
Anthropic lists Claude Opus 5.5 at $4 input and $20 output per million tokens, with cache reads at $0.20 and cache writes at $5. Claude Opus 5 was $5, $25, $0.50 and $6.25.
Does an AI model price cut halve my chatbot or automation bill?
Only the model usage part can fall, and only if you are billed by actual usage or the vendor passes it on. Workflow platform plans, hosting, support and management fees stay the same.
How much does AI automation cost a small business per month after the cut?
It depends on your token volume and your tools. In the worked example, 1,000 lead replies come to $5 of GPT-6 Sol use, while Make's Core plan is $12 and Zapier Professional starts at $19.99. Your own invoices give the real figure.
How do I find out which model my automation uses?
Check the provider's usage reports or ask your vendor for the model identifier and tier. OpenAI's organization usage data can group by model and show requests and token counts.
Should I switch to the new models right away?
Test first. Run real inputs through the new model, check the quality and the token use, then compare the whole monthly bill. A small model line rarely justifies a rebuild on its own.
The Bottom Line
On September 22, 2026, OpenAI cut its newest API model prices by half against GPT-5.6, and Anthropic priced Claude Opus 5.5 below Opus 5. For a 1,000-reply automation, that turns a $10 model line into $5 on Sol. The workflow platform, hosting and support around it don't move. Your bill drops only where you pay for model usage directly or your vendor passes it on.
Once you know which line is which, you can ask your vendor for a specific number instead of a general discount, change models on purpose and after testing, and read the next price announcement in a few minutes rather than wondering. That puts you in charge of the AI fee instead of just paying it.
If you'd like a second set of eyes on the numbers, we at Web Leveling can trace your model and platform charges across accounts, test a model change against your real inputs, and set up usage reports you can read yourself. Our AI automation work keeps the model, the tools and the accounts clearly separated on the bill, and if the check shows the model line is too small to bother with, we'll tell you to leave it alone. We work with small and medium businesses across the country and overseas. Send us your current AI automation setup, and we'll help you see where the money goes.
Terms
AI pricing words in this post
Tap a term to see what it means.
Token. A small piece of text, often part of a word, that AI models count and bill by.
Input tokens. The text your automation sends to the model, such as a lead's message and your business details.
Output tokens. The text the model writes back, usually priced higher than input.
Cached input. Input the provider has stored from an earlier request, billed at a separate and often lower rate.
Service tier. The speed and pricing option a request runs on, such as Standard, Batch, Flex or Fast.
Pass-through. A billing setup where a vendor charges you the provider's usage cost, with or without a stated markup.
Tasks and credits. The usage meters Zapier and Make use to charge for each step a workflow runs.
