How Macha Credits & Billing Work (Per-Model, Per-Response)

That distinction matters more than it sounds, because most of the AI-support market prices the other way. This post walks through how Macha credits actually work: what one costs, why the same task can cost 1 credit or 3 depending on the model, how plans and top-ups fit together, and how to forecast a monthly bill without guessing.

Macha is the AI agent layer on top of the helpdesk you already run (Zendesk, Freshdesk, Gorgias, Front), so credits meter the agents' thinking and actions across your stack. Let's get into it.

A credit is one AI response

Every time one of your agents produces a response — drafts a reply, triages a ticket, calls a tool, decides a route — that turn consumes credits. The unit is the response, sometimes called an action or a message, depending on where you see it in the dashboard. It is deliberately not a "resolution" or a "deflection."

Why does Macha refuse to bill per resolution? Because Macha is automation and orchestration, not a black box that promises a fixed success rate. An agent might read a ticket, check Stripe, post an internal note, and hand off to a human — four real actions, zero "resolutions" by a strict definition, but genuine work that moved the ticket forward. Billing per action keeps the meter honest and, crucially, under your control: the levers that change your bill are how many responses your agents generate and which model they generate them with. Both are things you set.

There's one free exception worth knowing: the welcome chat a new org lands in after onboarding is on the house — a "This chat costs no credits" badge sits right next to the model picker. One per organisation, so you can kick the tyres before a single credit moves.

Per-model pricing: why 1 credit or 3

Here's the part that trips people up. The same response can cost a different number of credits depending on which model produced it. That's not a gotcha — it's the most useful lever you have. Bigger, more capable models cost more credits per response; fast, efficient models cost less. You match the model to the job.

As of the June 17, 2026 pricing update, the OpenAI lineup looks like this:

Across the full lineup — including the Claude and Groq models Macha also offers — per-response cost spans roughly 0.5 to 9 credits by model, with GPT-5.4 Mini at 1 credit as the default for most workflows. When Macha cut GPT-5 and GPT-5.4 costs in June, the practical effect was that Professional orgs got ~67% more GPT-5 and GPT-5.4 responses per month inside the same 10,000-credit allowance — same plan, more work.

Model choice is per conversation by default, with an org-wide default you set in Settings. Change it from the top-right dropdown and Macha shows a toast reminding you the switch applies only to the current conversation — if you wanted to move the whole org, it points you to Settings. The agents table also shows each agent's model in its own column, so you can scan model selection across your full fleet at a glance and spot anywhere you're paying for GPT-5.4 on a job GPT-5.4 Mini would handle.

A quick worked estimate

Say a triage agent on GPT-5.4 Mini handles 4,000 ticket events a month, one response each: that's 4,000 credits. A more involved resolution agent on GPT-5 answers 1,500 tickets, averaging two responses each: 1,500 × 2 responses × 2 credits = 6,000 credits. Total: 10,000 credits — exactly a Professional plan's monthly allowance (10,000 credits). The point isn't the precise figures; it's that you can do this arithmetic before you commit, because the unit is a countable action with a known per-model price.

How plans and credits fit together

Each plan bundles a monthly credit allowance that resets every billing cycle:

Plan tier also gates capacity — integrations, knowledge sources, agents, scheduled triggers, and features like Studies — not just the credit count. The current numbers live on the pricing page; annual billing knocks off a discount. Enterprise runs on bypassed credit checks rather than a fixed allowance, which is the right fit once volume is high enough that counting credits stops being useful.

Top-up packs: credits that don't expire

Monthly allowances reset to zero each cycle. So what happens on a spiky month — a product launch, a holiday rush, a Black Friday? You buy a top-up pack.

Macha offers three one-time top-up packs (1,000 / 5,000 / 10,000 credits), purchased from the billing page through an in-page Chargebee checkout. Two properties make them genuinely useful:

• Top-ups never expire. Unlike your monthly allowance, top-up credits roll over month to month until you spend them.
• They're spent automatically, second. Macha always burns your monthly allowance first, then dips into top-ups — so you never have to think about which bucket you're spending from, and you never lose a top-up credit to a reset.

On the Billing Usage card, the two buckets are shown separately: monthly usage gets a bar that resets each cycle, and a green +N top-up chip sits above it when you're carrying a top-up balance, with a hover that explains the persistence rule. Top-ups are available to active Starter and Professional subscriptions; Trial users upgrade to a paid plan first, and Enterprise plans don't need them.

This is a real, quiet differentiator. Several credit-based competitors reset unused credits to zero each month — buy for a busy month that never comes and the credits are simply gone. Macha's top-ups don't evaporate.

Staying ahead of the meter: usage alerts

You shouldn't discover you've run out of credits from a stalled agent. Macha fires credit usage alerts at 50%, 80%, and 90% of your monthly allowance. The alerts are deduplicated in the database (so a deploy can't re-send them) and fire once per threshold crossing with a single admin notification — not one email per admin in the org. When you cross 80%, that's your cue to either tune a chatty agent down to a cheaper model or buy a top-up before month-end.

How Macha's model compares (and where it doesn't fit)

It's worth being honest about the trade-off, because per-response billing isn't automatically "cheaper" than the alternatives — it's different, and the difference cuts both ways.

The rest of the market mostly meters per resolution: Intercom's Fin bills $0.99 per resolution (June 2026), Zendesk's AI agents run around $1.50+, and Gorgias is about $0.90 — you pay only when the AI "resolves" something. HubSpot uses a credit model closer to Macha's (≈50 credits per resolved conversation) but, again, resets unused credits monthly. (Competitor figures are vendor-published headline rates and exclude platform/seat fees; verify against current pricing pages.)

When per-response billing is the better deal: when your agents do a lot of non-resolution work — triage, routing, internal notes, tool calls, summarisation, Studies analysis — that a per-resolution vendor either can't bill for or bills as a full resolution anyway. And when you want a predictable, forecastable bill you control by choosing models, instead of one that floats with a "resolution rate" you can't know on day one.

Where it's less flattering — the honest watch-outs:

• A chatty multi-turn agent costs more than a one-shot deflection. If a conversation takes six model turns to resolve, that's up to six responses metered. Per-resolution pricing would charge once. Tighten agent instructions so they don't ramble, and reserve the expensive models for the steps that need them.
• You have to right-size models yourself. The flip side of control is responsibility: leaving a high-volume triage agent on GPT-5.4 when GPT-5.4 Mini would do is paying 3× for no benefit. The model column on the agents table exists precisely so you can catch this.
• Studies and bulk runs draw from the same balance. A large Studies analysis is credits too — that's why Studies shows a cost estimate and a confirm gate before any run. Plan for it.

If your support is almost entirely simple, one-touch deflections and you genuinely never want to think about model selection, a flat per-resolution tool may feel simpler. Macha's model rewards teams who want to see — and steer — exactly where their AI spend goes.

FAQ

What is one Macha credit? The cost of one AI response from an agent — a reply, a triage decision, a tool call. It is billed per action, not per resolved or deflected ticket.

Why does the same task cost a different number of credits? Because credits are priced per model. A response on GPT-5.4 Mini costs 1 credit; the same response on GPT-5 costs 2, and on GPT-5.4, 3. You choose the model per conversation, with an org default in Settings.

Do unused credits roll over? Monthly plan credits reset each billing cycle. Top-up credits never expire — they roll over until spent and are used automatically after your monthly allowance runs out.

What happens if I run out of credits mid-month? You'll have seen alerts at 50%, 80%, and 90% first. Buy a top-up pack from the billing page (it applies immediately) or upgrade your plan. Enterprise plans run on bypassed credit checks.

Does Macha charge per resolution like Intercom Fin? No. Macha bills per AI response, which keeps cost tied to actual work and under your control via model choice — rather than to a "resolution" definition the vendor sets.

How do I estimate my monthly credit usage? Multiply expected responses per agent by that agent's per-model credit cost, and sum across agents. The pricing page and the Billing Usage card help you sanity-check against real usage.

Start metering your own agents

The fastest way to understand Macha's credits is to watch the Usage card move on your own tickets. Start a 7-day free trial, no credit card required, connect your helpdesk, and the welcome chat is free — see the billing docs for the full breakdown, or read how Studies uses the same credit balance to analyse your back-catalogue of tickets.

Written by Abbas (Customer Support & AI, Macha) · Reviewed by Ankeet Guha (Co-founder & CTO) · Published 2026-06-24 · Last updated 2026-06-24.