Short answer: the AI inside your chatbot, lead form or invoice reader runs on a model you rent from someone else, and that model's price and lifespan are set by them. Google's pricing page shows its two current Gemini Flash models doubling in price on 1 January 2027, and both Google and OpenAI have older models being rerouted or shut down between now and January. If any of your tools rely on these, December is the month to check.
What's changing, with dates
Three changes are already on the vendors' own pages.
Gemini Flash prices double in January. Google's Gemini API pricing page lists Gemini 3.8 Flash and Gemini 3.6 Flash at $0.75 per million input tokens and $3.75 per million output tokens at the standard paid-tier rate "through December 31, 2026". From 1 January 2027 both become $1.50 and $7.50. Context caching doubles on the same date. The free tier stays free, though the same table notes that free-tier usage is used to improve Google's products and paid usage is not.
Gemini 2.5 calls get moved for you on 1 December. Google's Firebase AI Logic model page says that from 1 December 2026, Gemini Developer API requests to the general-use Gemini 2.5 Flash and Pro models "will be automatically routed to newer models", which "may cause breaking changes". Its table shows 2.5 Pro requests going to gemini-3.8-flash, and 2.5 Flash requests going to gemini-3.5-flash-lite. So a tool that was set up on a Pro model could find itself answering from a Flash one.
Several OpenAI models shut down in December and January. OpenAI's deprecations page lists gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3 and o3-pro for shutdown on 11 December 2026, and the listed dated versions of GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini on 14 January 2027. Each has a recommended replacement. OpenAI says affected customers are told by email and in its documentation.
None of this is unusual. OpenAI's own page says it regularly retires older models. What catches businesses out is that nobody inside the company is reading those emails.
Why it lands on an owner's desk
When a team builds a chatbot or an automation, the model name usually sits in one line of code or one setting in a tool. The person who chose it may have left, or it may have been an agency. The billing email might go to a founder's inbox or a card statement nobody reads line by line.
Two things can then happen without anyone touching the tool.
The bill changes. Take a made-up example: an enquiry assistant handling 1,000 chats a day, each using about 3,000 input tokens and 500 output tokens, on Gemini 3.8 Flash. That's roughly 90 million input and 15 million output tokens a month, or about $124 at the current listed rates and about $248 at the January rates. Small for one tool, but these add up across a few of them, and a jump like that in a month's card statement deserves an explanation in advance.
The answers change. A rerouted or replaced model doesn't behave exactly like the old one. It may format replies differently, follow your instructions a little more or less closely, or handle Hindi and Hinglish differently. Firebase's documentation also warns against "-latest" model aliases for this reason: they get swapped with every new release, and a notice email two weeks ahead is sent only for breaking changes.
A 45-minute check for December
Do this once, with whoever looks after your AI tools, inside or outside the company.
- List every place AI runs on your behalf. Chatbots, lead scoring, invoice reading, call summaries, report builders. Include anything an agency built for you.
- Write down the model and vendor next to each. If the answer is "the latest one" or "not sure", that's the first thing to fix. A specific model version is easier to plan around than an alias.
- Check each one against the three pages above. Is it a Gemini 2.5 model, a Gemini Flash model on the paid tier, or an OpenAI model with a shutdown date? Note the date.
- Find out who receives the vendor's emails. Point that address at someone who reads it, or forward it to a shared inbox the team watches.
- Look at last month's usage, then do the January sum. Your vendor's usage dashboard or monthly bill shows what you spent. Doubling the Gemini Flash share gives you a rough January figure.
- Set a budget alert, and know what it does. Google Cloud's budget documentation is clear that an alerts-only budget "doesn't automatically cap" usage or spending. It warns you, and someone still has to act.
- Test the replacement before the date. Run 20 or 30 of your real cases through the new model and compare. We wrote up how to test a model on your own work on 29 October 2026.
If a tool came from a vendor, ask them three questions: which model it uses, whether a price change will be passed on to you, and how they test a replacement before switching.
When the price goes up, look at the size
A price rise is a good prompt to ask whether the tool needs that model at all. Some everyday jobs, like sorting enquiries or pulling fields from a form, can run on smaller, cheaper models. Google's own table names gemini-3.5-flash-lite as the replacement for Gemini 2.5 Flash. Others may justify paying for the bigger model.
The answer will differ tool by tool, and it should come from testing on your own work, not from a leaderboard. What matters is that the decision is yours, made in December, and not something that happens to you on 1 January.
Arthat AI
Working on something similar at your company?
Arthat AI builds these automations for Indian businesses every week. If you want to talk through whether this fits your situation, we'll give you a straight answer — even if the answer is "don't build it."


