New AI models this autumn: what they change for your chatbot
GPT-6 Sol and Luna, Claude Opus 5.5 and Sonnet 5.5, Gemini 3.8 Flash: what a small FAQ chatbot costs a month on each, and why a small model is usually enough.
Nikolay Tolmachev, Founder, Tolman Studios
6 min readNews
In short
$0.10per million input tokens for GPT-6 Luna, the smallest new OpenAI model (OpenAI, September 2026)
- In September OpenAI released GPT-6 Sol and Luna at half the API price of the models before them. Anthropic released Claude Opus 5.5 (20% cheaper than Opus 5) and Sonnet 5.5 (same price as Sonnet 5).
- By my estimate, a typical FAQ bot with 1,000 conversations a month costs about $2-14 in model usage on a small model and $36-72 on a large one.
- For answers about opening hours, prices and services, a small fast model is usually enough. The chat on this site runs on Gemini 3.5 Flash-Lite.
On this page
September brought a new AI model almost every week. OpenAI released GPT-6, Anthropic released Claude Opus 5.5 and Sonnet 5.5, and Google moved its Gemini Flash to version 3.8. Most of them are cheaper than the models they replace.
If you run a chatbot that answers customer questions, here is the short version. The big new models are impressive, but an FAQ bot probably doesn't need them. Below I work out the monthly cost on six models.
What came out in September?
2 Sep 2026
Google makes Gemini 3.8 Flash generally available. Its price of $0.75 per million input tokens is an introductory one, valid until 31 December 2026.
3 Sep 2026
OpenAI launches GPT-6 Astra, its most capable model. Access opens in phases: first a cybersecurity programme, then paid ChatGPT plans and the API.
22 Sep 2026
Anthropic releases Claude Opus 5.5, 20% cheaper than Opus 5. The same day OpenAI releases GPT-6 Sol and GPT-6 Luna at half the API price of the 5.6 versions.
28 Sep 2026
Anthropic releases Claude Sonnet 5.5 at the same price as Sonnet 5. Anthropic says it needs fewer tokens per task.
6 Oct 2026
Today. The prices in this post are the ones on the providers' own pages.
For a small business, two things matter here. Prices keep falling. And each company now has a clear "small and fast" model next to the big one. GPT-6 Luna is OpenAI's model for high-volume, clearly defined tasks: quick answers, summaries, pulling facts out of text. That describes most FAQ bots.
What do the new models cost?
AI models are paid per token. A token is a small piece of a word. The providers charge separately for what the model reads (input) and what it writes (output). All prices are in US dollars per million tokens, standard API, as of 6 October 2026.
| Model | Company | Input | Output | Made for |
|---|---|---|---|---|
| Claude Opus 5.5 | Anthropic | $4 | $20 | The hardest tasks |
| Claude Sonnet 5.5 | Anthropic | $2 | $10 | Everyday work, coding, agents |
| GPT-6 Sol | OpenAI | $2 | $10 | Complex tasks, coding |
| Gemini 3.8 Flash | $0.75 | $3.75 | Fast general use (intro price until 31 Dec 2026, then $1.50 / $7.50) | |
| Gemini 3.5 Flash-Lite | $0.30 | $2.50 | Cheap, high-volume tasks | |
| GPT-6 Luna | OpenAI | $0.10 | $0.50 | High-volume, clearly defined tasks |
The gap between the top and the bottom of this table is 40 times on both input and output.
What does an FAQ bot cost a month on each model?
Prices per million tokens mean little until you put a real bot behind them. So here is my estimate for a typical small-business FAQ bot. These are my assumptions, not anyone's published data:
- 1,000 conversations a month, about 33 a day.
- 6 messages per conversation: 3 questions from the customer, 3 answers. That makes 3,000 answers a month.
- Each answer, the model reads about 4,500 tokens: your instructions and FAQ (about 4,000) plus the conversation so far.
- Each answer is about 300 tokens long, including a little "thinking".
- No discounts for caching. That keeps the estimate on the safe side.
That is 13.5 million tokens read and 0.9 million written per month.
- Claude Opus 5.5$72
- Claude Sonnet 5.5$36
- GPT-6 Sol$36
- Gemini 3.8 Flashintro price until 31 Dec 2026$13.50
- Gemini 3.5 Flash-Literuns the chat on this site$6.30
- GPT-6 Luna$1.80
My estimate with the assumptions above. Prices: Anthropic, OpenAI and Google API price pages, checked 6 October 2026
For a small bot, all six are affordable. On a small model, the platform subscription and the setup usually cost more than the model itself. The choice matters more when the volume grows. At 10,000 conversations a month, Opus 5.5 would cost about $720 and Luna about $18.
There is one thing the chart can't show. Large models often think longer before they answer, and you pay for that thinking as output. In real use the gap between big and small is often wider than in my table.
Do you need the newest big model?
A small fast model is enough if
- The bot answers from one fixed text: services, prices, hours, address
- Most questions repeat every week
- Unclear cases go to a person anyway
- You want answers in a second or two
A bigger model is worth testing if
- The bot has to compare several options and explain a choice
- It works through long documents, contracts or price lists that change often
- It takes several steps on its own, like checking a calendar and then booking
- The small model gets real questions wrong in your tests
Which model runs the chat on this site?
The chat on this site runs on Gemini 3.5 Flash-Lite, on a paid Google account, with thinking set to the minimum. It answers from one text about my services, prices and process, in six languages.
A few days ago my first version ran on Claude Opus 5.5. Then I did the same calculation for my own traffic and moved to the small model. For this job the large one was mostly extra cost. I'll keep reading the answers. If the small model starts to get things wrong, I'll test a bigger one.
That is how I would decide for a client too. Start small and test on real questions. Move to a bigger model only when the answers show you need it.
What should you do now?
- Look up which model your chatbot uses. Ask your provider if it's not in the settings.
- Take 20-30 real questions from last month's chats.
- Run them through your current model and one cheaper model. Compare the answers side by side.
- Switch only if the cheaper model is as good, or the bigger model fixes real mistakes.
- If you use Gemini 3.8 Flash, put a reminder in your calendar for December. Its intro price ends on 31 December 2026.
The full picture of what a chatbot costs, with platform fees and WhatsApp, is in my post on how much an AI chatbot costs a small business. If you are still deciding whether you need a bot at all, start with where AI chatbots help and where they don't. And if you want me to pick the model and set the bot up, here is my AI and automation work.
Questions people ask
Should I switch my chatbot to GPT-6 or Claude 5.5?
Only if the bot gets real questions wrong today. Test the new model on 20-30 real questions from your chats first. A newer large model costs more per answer, and a simple FAQ bot rarely needs it.
How much does the AI model in a small chatbot cost per month?
In my example of 1,000 conversations a month, about $2 on GPT-6 Luna, $6 on Gemini 3.5 Flash-Lite and $72 on Claude Opus 5.5. It is an estimate with stated assumptions. Your numbers depend on how long your texts and conversations are.
Is the model the biggest part of a chatbot bill?
For a small business, usually not. The platform subscription, the setup and the care often cost more than the model itself, especially with a small model.
Do the prices stay like this?
No guarantee. Google already says Gemini 3.8 Flash will cost twice as much from 1 January 2027. Check the price page of your provider once a quarter.
Sources
- Introducing Claude Opus 5.5, Anthropic, checked 6 October 2026
- Introducing Claude Sonnet 5.5, Anthropic, checked 6 October 2026
- Claude API pricing, Anthropic, checked 6 October 2026
- API pricing, OpenAI, checked 6 October 2026
- OpenAI launches GPT-6 Sol and Luna, TechCrunch, 22 September 2026
- OpenAI launches Astra, its powerful and controversial new model, TechCrunch, 3 September 2026
- Gemini Developer API pricing, Google, checked 6 October 2026
- Gemini API release notes, Google, checked 6 October 2026
Facts and prices are as of the dates in the sources.
Nikolay Tolmachev
Founder, Tolman Studios
I build websites, AI chatbots and ads for small businesses, working from Slovakia. Before that I spent over 4 years managing IT projects. I hold PMP and Certified ScrumMaster certificates.