Skip to content

Message credits: how Chatsy charges for AI replies

Each AI reply costs 1 to 30 credits depending on the model. See credits per model, what each plan covers, when replies stop and how to estimate a month.

Updated

Chatsy bills AI work in message credits. Every time an agent writes a reply, Chatsy deducts the credit rate of the model that agent uses: 1 credit for budget models, up to 30 for the most expensive model on the Pro plan. Seats, conversations and resolutions are not what you pay for. A plan gives your workspace a fixed number of credits each month, and the model you pick decides how many replies those credits buy.

What uses credits

A credit is deducted when an AI model generates a response to a visitor or to you in the playground. The cost is the model's rate multiplied by the number of responses, so a 2-credit model uses 2 credits per reply.

These do not use credits:

  • A teammate typing a reply in Live chat or the Inbox.
  • Visitors opening the widget, browsing suggested questions or leaving an email for follow-up.
  • Adding or re-reading knowledge sources.

AI drafts (suggested replies for tickets and live chat) are a separate module. The Modules page states that each draft uses message credits, so include drafts in your estimate if you turn that module on.

Credits per model

The rate is set per model, not per plan. A model only appears in an agent's model list if your plan includes it.

ModelProviderCredits per replyLowest plan
Claude Haiku 5.5Anthropic1Free
Gemini 2.5 Flash LiteGoogle1Free
Gemini 3.1 Flash LiteGoogle1Free
GPT-4o MiniOpenAI1Free
GPT-6 LunaOpenAI1Free
DeepSeek ChatDeepSeek1Hobby
DeepSeek V3.2DeepSeek1Hobby
DeepSeek V4.1 Flash 1017DeepSeek1Hobby
Gemini 2.5 FlashGoogle1Hobby
GPT-4.1OpenAI1Hobby
GPT-4.1 MiniOpenAI1Hobby
GPT-4.1 NanoOpenAI1Hobby
GPT-4oOpenAI1Hobby
GPT-5 MiniOpenAI1Hobby
GPT-5 NanoOpenAI1Hobby
GPT-5.2 ChatOpenAI1Hobby
Llama 3.1 70BMeta1Hobby
Mistral Large 3Mistral1Hobby
Mistral Large 4Mistral1Hobby
Claude Haiku 4.5Anthropic2Hobby
Claude Sonnet 5.5Anthropic2Hobby
Gemini 2.5 ProGoogle2Hobby
Gemini 3.5 FlashGoogle2Hobby
Gemini 4 ArgonGoogle2Hobby
GPT-5OpenAI2Hobby
GPT-5.2OpenAI2Hobby
GPT-5.2 CodexOpenAI2Hobby
GPT-6.1 SolOpenAI2Hobby
o4 MiniOpenAI2Hobby
Claude Sonnet 4.5Anthropic3Hobby
Claude Sonnet 4.6Anthropic3Hobby
GPT-5.3 CodexOpenAI3Hobby
o3OpenAI3Hobby
Claude Opus 5.5Anthropic4Hobby
Claude Opus 4.6Anthropic5Hobby
Claude Opus 4.8Anthropic5Hobby
Claude Fable 5Anthropic10Hobby
GPT-6 AstraOpenAI10Hobby
Llama 3.3 70B InstructMeta1Pro
Claude Opus 4.1Anthropic15Pro
GPT-5.2 ProOpenAI30Pro
o3 ProOpenAI20Enterprise

Older model IDs that providers have retired keep working for existing agents: Chatsy maps them to a current model and charges that model's rate.

What each plan covers

Plans differ in monthly credits and in which models you can choose. Hobby and Standard share the same model list, so the highest rate on those plans is 10 credits. Pro adds models at 15 and 30 credits, and Enterprise adds one more.

PlanModels availableHighest credits per reply
Free51
Hobby3810
Standard3810
Pro4130
Enterprise4230

The table below turns each allowance into replies. The figures already allow for the cap described in the next section.

PlanPriceCredits per monthReplies at 1 creditTop model rate (credits)Replies at top rateTeam seats
Free$0/mo40391391
Hobby$40/mo50049910491
Standard$150/mo4,0003,999103993
Pro$500/mo15,00014,999304995
EnterpriseQuote100,00099,999303,33310

Annual billing costs ten times the monthly price and gives the same monthly credits. Seats include the workspace owner, so a 1-seat plan has no room to invite teammates.

When replies stop

Chatsy checks the allowance before it sends each AI reply. If the reply would bring the month's usage to or past the allowance, Chatsy blocks it and returns this error instead of an answer:

Credit limit reached. Your {Plan} plan allows {N} credits per month.

Because a reply that would land exactly on the allowance is blocked, the last credit of the month cannot be spent. That is why Hobby's 500 credits cover 499 one-credit replies, not 500.

When a workspace reaches its limit you have two options: upgrade the plan from Settings → Billing, or wait for the monthly reset. Chatsy does not sell credit top-ups or charge overage fees, and unused credits do not carry over.

Teammate replies in Live chat and the Inbox never use credits, so your team can keep answering while AI replies are paused.

When credits reset

Credits reset once a month. The reset happens on the first credit check after the reset date passes, and the next reset date is set one calendar month later. It is not tied to your billing date, so check the usage figure in the dashboard rather than counting from your invoice.

How to estimate a month

Work it out before you choose a plan:

  1. Count the conversations you expect the agent to handle in a month.
  2. Multiply by the average number of AI replies per conversation. Look at a week of real chats: many support conversations need two or three replies, not one.
  3. Multiply by the credit rate of the model the agent will use.
  4. Add AI drafts if you will turn that module on.
  5. Pick a plan whose allowance stays above the total, with room for busy weeks.

Worked example. A store expects 600 conversations a month, with 2.5 AI replies per conversation on average, so about 1,500 replies.

Model choiceCredits per replyCredits neededSmallest plan that covers it
A 1-credit model11,500Standard (4,000 credits)
A 2-credit model23,000Standard (4,000 credits)
A 10-credit model1015,000Pro covers 1,499 replies at 10 credits, so it falls just short. Use a cheaper model or split the volume.

Hobby's 500 credits would not cover this store with any model: 499 one-credit replies is a third of the 1,500 it needs.

The estimate at the top of the pricing page runs the same check for every plan.

Choosing a model with credits in mind

Start with the cheapest model that answers your test questions correctly. Then try a stronger model only on the questions it gets wrong. If the cheaper model passes, every reply costs less and the same allowance covers more conversations. You can change the model per agent at any time in the agent's Settings → AI tab, and the new rate applies from the next reply.