Raw AI model costs fell sharply in the past three weeks. Your automation bill is a different story. Here is what is verifiable, and how to check whether your tools pass any savings along.
According to press coverage of OpenAI's September 22 announcement, GPT-6 Sol dropped from $4 to $2 per million input tokens and from $20 to $10 per million output tokens. The smaller Luna model reportedly went from $0.20 to $0.10 input and from $1.20 to $0.50 output. Reports differ on whether the old prices were a promotional baseline, and we could not confirm the new rates are permanent, so check OpenAI's pricing page before budgeting.
On October 7, Anthropic launched Claude Haiku 5.5, aimed at narrow, high-volume tasks. Reported API pricing is $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, with higher rates above that. The figures come from a secondary roundup, so verify with Anthropic.
The direction is clear: the cost of routine AI work like classifying an email, extracting invoice fields or summarizing a call is falling at the model layer.
Token prices are what developers pay. Automation platforms sell something else, and they price it their own way.
Make bills in credits, where a standard module run typically counts as one credit, and its recent Maia assistant draws from the same credit pool on paid plans. Zapier bills in tasks, and our earlier coverage of its AI-step changes covers how model tier can affect task counts. In both cases the price you see is a bundle of platform usage and AI usage, not a pass-through of token costs.
We did not find an announcement from Zapier, Make, n8n or HubSpot lowering AI-step pricing because of these model price cuts. That could change, and vendors may adjust quietly, so treat this as a point-in-time observation.
| Setup | Do model price cuts reach you? |
|---|---|
| Built-in AI steps on a platform plan | Only if the vendor changes its task or credit rules. Check your plan and usage page. |
| Your own API key inside Make or Zapier | Likely yes for the model portion, since you pay the provider directly. You still pay the platform for each run. |
| Self-hosted n8n with your own key | Yes for the model portion. Hosting and upkeep are on you. |
| AI inside a CRM such as HubSpot | Depends on the vendor's own credit system, which is set separately from model prices. |
Compare options in our Make vs Zapier breakdown and the Make Maia credits explainer.
1. Find your AI-heavy workflows. Look at your usage page and note which automations run the most AI steps. High-volume, simple jobs like tagging, routing and extraction are where cheaper models matter most.
2. Ask whether the step needs a premium model. OpenAI's own guidance, as reported, points to the smaller model for summarizing and extraction and the larger one for complex reasoning. If your platform lets you choose the model, test the cheaper option on a small batch with sample data first.
3. Price both routes. Estimate your monthly runs, then compare the platform's built-in AI step against bringing your own key. Include the platform fee for each run in both cases. Do not assume the own-key route wins until you have run the numbers.
4. Watch annual contracts. If you are about to prepay for a year of any AI-enabled tool, ask the vendor whether lower model costs will be reflected in pricing.
Cheaper models are good news, but most small businesses will not feel them automatically. The savings show up when you control the model call, or when a vendor decides to pass them on. Until then, the useful move is to measure how many AI steps your automations actually run, and keep an eye on vendor pricing pages.
Note that StackPilot's comparisons are based on research of public information and vendor documentation, not hands-on testing. Model pricing here comes from secondary reports and may be out of date; confirm with each provider.