Account & Billing
Credits and model costs
How Gabbex credits work, how much each AI model costs per reply, and how to make a monthly allowance go further.
Every AI reply your assistant sends costs credits. Credits are how Gabbex measures usage across models that have very different underlying costs — a single accounting unit that lets you mix and match models without thinking about per-token billing.
This page explains what a credit is, how much each model costs, and how to choose a model that fits your monthly allowance.
What a credit is
A credit is the cost of a single AI reply from your assistant. When a visitor asks a question and the assistant responds, the model’s credit cost is deducted from your workspace’s monthly credit balance.
The credit cost depends on which model the assistant is using:
- 1 credit — fast, lightweight models. Best for high-volume support and most everyday questions.
- 2 credits — mid-tier models with stronger reasoning. Best for nuanced sales conversations and complex support.
- 3 credits — top-tier reasoning models. Best for difficult or high-stakes questions where quality is more important than cost.
Credits do not charge per token, per message length, or per conversation. They charge once per assistant reply, with a flat cost determined by the model. This makes the meter predictable.
What counts as a credit-charged reply
Charged (1 credit at the model’s credit_cost, once per reply):
- Every successful AI message the assistant sends to a visitor — the answer to a question, a follow-up, or the message that asks for a name and email during lead capture.
- Test chats from the dashboard Chat tab. Test chats use the exact same chat pipeline as live conversations and are charged the same way.
- Replies that involve a tool call. When the assistant decides to call lead capture, escalate-to-human, or a Shopify lookup, the internal back-and-forth with the tool happens inside a single completion. The visitor sees one final assistant message, and you are charged once for it — never twice.
Not charged:
- The visitor’s own messages. Only AI output is metered. Industry-standard chatbot platforms work the same way.
- Failed completions. If the model fails to produce an answer and the assistant returns its built-in fallback (“I’m sorry, I couldn’t generate a response.”), no credit is consumed.
- Indexing work. Crawling a website, syncing Notion, or uploading a file is separate from credits and does not draw from your monthly allowance.
- Token volume. Credits are flat-per-reply, not per token. A 30-word answer and a 300-word answer cost the same number of credits as long as they come from the same model.
Monthly credit allowance by plan
Each plan includes a monthly credit allowance for the entire workspace:
| Plan | Monthly credits |
|---|---|
| Free | 50 |
| Spark | 1,000 |
| Core | 4,000 |
| Scale | 10,000 |
Credits are pooled across every assistant in the workspace. If you have two assistants on the Core plan, they share the same 4,000 credits per month — you do not get 4,000 each.
Credits reset at the start of each billing period. Unused credits do not roll over.
What this gets you in practice
How many AI replies a credit allowance buys depends entirely on which model your assistants are running on. Three quick examples for a workspace on Core (4,000 credits / month):
- All-1-credit setup (default GPT-4o Mini, Gemini Flash, Claude Haiku): up to 4,000 AI replies per month.
- All-2-credit setup (GPT-5 family, Gemini Pro, Claude Sonnet): up to 2,000 AI replies per month.
- All-3-credit setup (Claude Opus): up to ~1,333 AI replies per month.
The math is the same on every plan: divide your monthly credit allowance by the model’s per-reply credit cost.
Most teams mix models — for example, a high-traffic support assistant on a 1-credit model and a lower-volume sales assistant on a 2-credit model. Because credits are pooled across the workspace, the total still has to fit inside the plan’s monthly allowance.
Reminder: only AI replies consume credits. Visitor messages, failed completions, and indexing work are free. See What counts as a credit-charged reply above.
Model access by plan
Not every plan can use every model:
- Free plan — only the default model (GPT-4o Mini, 1 credit per reply). You cannot switch to a different model on the free plan.
- Spark, Core, and Scale — full access to every model in the catalogue. Switch models per assistant from General → Model.
If you want to try a Claude or Gemini model, you need to be on at least the Spark plan.
Model catalogue and credit costs
The full list of models you can pick from on a paid plan, grouped by credit cost.
1 credit per reply — fast tier
Best for high-volume customer support, FAQs, and the majority of everyday questions. These models are the cheapest to run and have the lowest latency.
| Provider | Model |
|---|---|
| OpenAI | GPT-4o Mini (default) |
| OpenAI | GPT-5 Mini |
| OpenAI | GPT-5.4 Mini |
| OpenAI | GPT-5.4 Nano |
| Gemini 2.5 Flash | |
| Gemini 3 Flash | |
| Gemini 3.1 Flash Lite | |
| Anthropic | Claude 4.5 Haiku |
GPT-4o Mini is the default for new assistants and the only model available on the Free plan. It is also the model most teams should stay on — it is fast, cheap, and accurate enough for the vast majority of customer questions.
2 credits per reply — mid tier
Best for sales conversations, complex support, and assistants that need to reason across longer context. Slightly slower than the fast tier but noticeably stronger on nuance.
| Provider | Model |
|---|---|
| OpenAI | GPT-5 |
| OpenAI | GPT-5.1 |
| OpenAI | GPT-5.2 |
| OpenAI | GPT-5.4 |
| OpenAI | GPT-4o |
| Gemini 2.5 Pro | |
| Gemini 3.1 Pro | |
| Anthropic | Claude 4.5 Sonnet |
| Anthropic | Claude 4.6 Sonnet |
3 credits per reply — top tier
Best for the highest-stakes assistants — internal knowledge bases for senior teams, technical support for complex products, or any situation where you would rather pay more per reply for the best possible answer.
| Provider | Model |
|---|---|
| Anthropic | Claude 4.5 Opus |
| Anthropic | Claude 4.6 Opus |
Picking the right model for your assistant
A few rules of thumb:
- Start with the default (GPT-4o Mini, 1 credit). It handles most customer-facing use cases and lets you serve up to four times more replies than a 2-credit model on the same plan.
- Move to a 2-credit model only if you see weak answers in the Conversations tab that are clearly a model issue, not a knowledge gap. If the answer is missing because the source does not exist, switching models will not help — add a Q&A entry instead.
- Use a 3-credit model only when you have to. Opus models are roughly three times more expensive per reply than the default. They are the right call for niche, complex assistants — not for a typical Shopify or service-business assistant.
- Run different assistants on different models. A high-volume “Support” assistant on a 1-credit model and a low-volume “Sales” assistant on a 2-credit model is a common setup. Both share the same workspace credit pool.
You can change the model at any time from General → Model inside the assistant. The change takes effect on the next reply.
Watching your credit usage
Open the Usage page in the sidebar to see:
- Total credits consumed in the current billing period.
- A daily timeline so you can spot spikes (7d / 30d / 90d views).
- A By model breakdown so you know which LLM model is consuming most of your credits.
- A paginated history of individual usage records — one row per reply, with the assistant, model, and credits charged.
Your plan’s monthly credit allowance is shown on the Subscription page next to the credits used so far.
When you cross 80% of your allowance, a banner appears in the dashboard. When you hit 100%, the assistant stops sending new replies until either the next billing period starts or you upgrade.
What happens when you run out
If a visitor messages your assistant after the credit allowance is exhausted:
- The assistant cannot generate a reply.
- The conversation is recorded so you can see what was asked.
- An error is returned to the widget so the visitor knows the assistant is temporarily unavailable.
To restore service immediately, upgrade to a higher plan from the Subscription page. The new credit allowance is available the moment payment succeeds.
Reducing credit burn
If you are using more credits than expected:
- Audit which model is consuming most. Open Usage and check the By model breakdown. If a 2-credit or 3-credit model is at the top of the chart, ask whether the assistant on that model really needs it.
- Switch heavy-traffic assistants to a 1-credit model. Going from a 2-credit to a 1-credit model halves the burn for the same conversation volume.
- Tighten lead capture and escalation. Each tool-driven message also costs credits. If lead capture is asking too aggressively, lower its sensitivity (see Lead capture tool).
- Strengthen your sources. Better sources mean shorter, more direct conversations. Long back-and-forths because the assistant is fishing for clarification consume more credits than a single confident answer.
- Block off-topic chatter. Set the off-topic policy in General to politely decline so visitors who try to use the assistant as a generic chatbot do not eat your allowance.
Frequently asked questions
Do unused credits roll over? No. Credits reset at the start of each billing period.
Are credits per-assistant or per-workspace? Per-workspace. All assistants in a workspace draw from the same monthly pool.
Does the test chat in the dashboard cost credits? Yes. Test chats from the dashboard Chat tab go through the same pipeline as live conversations, so each AI reply is charged at the model’s normal credit cost. If you want to do a lot of testing without burning through your allowance, switch the assistant to a 1-credit model first.
Do I get charged separately by OpenAI, Google, or Anthropic? No. Gabbex covers all model provider costs. You only pay your Gabbex subscription, and the credit allowance is what controls how much usage that subscription includes.
What if I switch models mid-month? You can switch any time. The credit cost updates immediately for new replies. Replies already sent under the old model are not retroactively recharged.
Can I buy extra credits without upgrading? Not yet. The current way to add capacity is to upgrade to a higher plan. If this is blocking you, email support@gabbex.com.