Pricing · Credits

How credits work

Juma runs on one pool of credits per workspace. A credit meters the tokens each model call sends and receives, at a published weight for every model.

21 Aug 2026
rates last changed

What is a Juma credit?

A credit is the unit Juma uses to meter AI compute. Every chat reply, Flow step and agent action sends tokens to a model and receives tokens back, and each model has a published weight in credits per million tokens. One workspace holds one pool of credits, shared by every member, with unlimited seats. Plans differ only in how many credits they include each month; the tiers are on the pricing page.
Sparkles Icon

One pool per workspace

Every member draws on the same balance. Usage shows per member, model, tool and project.

Sparkles Icon

Unlimited seats

Adding a teammate changes nothing on the invoice. Running more work does.

Sparkles Icon

Weighted per model

Each model has a published weight in credits per million tokens. The table below is the ledger.

Model weights

What does each model weigh?

Each model carries a published weight in credits per million tokens, with input and output priced separately. The weights below apply to every workspace on a credit plan. Claude Opus weighs about 1.7 times Claude Sonnet. Gemini 3.1 Pro weighs about two thirds of Sonnet.

Model
Credits per 1M input tokens
Credits per 1M output tokens
Claude Opus 5
714
3,571
Claude Opus 4.8
714
3,571
GPT-5.5
714
4,286
GPT-5.6 Terra
446
2,676
Claude Sonnet 5
429
2,143
Claude Sonnet 4.6
429
2,143
Gemini 3.1 Pro
286
1,714
GPT-5.6 Luna
29
171
Light models (Claude Haiku 4.5, Gemini 3 Flash, Gemini 2.5), image generation, web search
Billed by consumption at smaller weights
 
Rates last changed 21 August 2026.

The table is a ceiling, not a forecast: prompt caching, context compression and model routing mean a workspace's effective consumption runs below a raw token calculation. GPT-5.6 Terra is the default model for new workspaces; workspace admins can change the default, and any member can pick a different model for a chat.

The weights are the same for every workspace. When a weight changes, the table changes with it and the date above moves. New models are added with their weight at launch. Changes to the credit price itself are announced to workspace admins by email in advance.
Worked examples

What does a Flow run actually consume?

A Flow run is a chain of model calls. Each call sends the conversation so far, the project's instructions and any tool results to a model, and gets text back. Credits count those tokens at the model's weight. Three real runs from our own workspace show the scale: 14 calls for a case study, 35 for an ads audit, 67 for a cross-channel performance analysis.

Measured in our own workspace, August 2026, each on a single pass with no revisions. Same model for all three, so the difference is the work: more calls, more context carried into each call, more data from connected tools, and a longer deliverable at the end. Integrations and MCP tools have no price of their own; the calls they trigger are metered like any other. Run the same Flow on Claude Opus 5 and the Opus weights apply instead. The same Flow costs more with a longer brief, deeper research, more revisions or a longer deliverable. Image generation, web search and light models bill by consumption at smaller weights. The usage page in workspace settings shows what each run cost, per member, model, tool and project.

Why do Flows use fewer credits than long chats?

A Flow starts a clean thread, carries a finished procedure in its opening prompt and runs straight to the deliverable. A long chat works the procedure out live, and every call re-carries the whole conversation. Three habits keep a workspace's spend predictable.

The project holds the what, the Flow holds the how, and they stack. Strategy, taste and judgment stay with the team.
Sparkles Icon

Save recurring work as a Flow

The trial-and-error gets paid for once, when the Flow is saved, not every time the task comes round.

Sparkles Icon

Start a fresh thread for a new task

A thread that has been running all week carries all week's context into every call.

Sparkles Icon

Keep project knowledge lean

Instructions and knowledge ride along on every call in the project. Ten focused pages beat a hundred loosely related ones, for quality and for credits.

What happens to unused credits?

Unused plan credits carry into the following month and expire at the end of it, on monthly and annual plans alike. Top-up credits last 12 months from purchase and survive plan changes. Referral credits do not expire. The balance does not accumulate across the year.

A light month is not lost: whatever is left in August is there in September. It does not stack beyond that. If the balance keeps carrying over, the plan is too big; if the team tops up every month, it is too small.
Sparkles Icon

Plan credits, monthly or annual

Valid for the billing month plus the following month.

Sparkles Icon

Top-up credits

Valid for 12 months from purchase. They survive plan changes.

Sparkles Icon

Referral credits

No expiry.

What happens when the balance runs out?

Nothing cuts off by surprise. The balance is visible all month. When it falls below 10% of the plan, every workspace admin gets an email. At zero, the workspace either pauses new AI conversations until credits are added, or tops itself up automatically with the amount you set.
Sparkles Icon

Top up anytime

Packs run from 1,000 to 30,000 credits, purchased in billing settings. Top-up credits last 12 months.

Sparkles Icon

Auto top-up

Set a threshold and an amount. When the balance drops below the threshold, Juma adds the pack and emails the admins. No conversation waits.

Sparkles Icon

Resize the plan

Plans move up anytime and down from the next billing cycle. A team that tops up every month is better off one plan size up.

FAQ

Questions, answered

Are credits per member or per workspace?
Per workspace. Every member draws on the same pool, and seats are unlimited on every plan. Admins can cap what one member spends per month, which keeps the pool predictable without charging anyone per person.
Do integrations or MCP tools cost extra credits?
No separate price. When a Flow pulls data from HubSpot, Google Analytics or an MCP server, the tokens that call sends and receives are metered at the model's weight, and the usage page attributes them to the tool. Your subscription to the third-party tool is separate and unchanged.
Do image generation and web search use credits?
Yes, by consumption and at smaller weights than the large text models. A web search step inside a Flow is metered like any other call. Generated images are metered by consumption. Both appear on the usage page under the tool that produced them.
Can we cap what one person spends?
Yes. Admins set a monthly credit limit per member in the usage settings. A member who reaches the limit sees it in the app and can ask an admin to raise it. The rest of the team is unaffected.
How do we estimate what our team needs?
Run the team on a plan for two weeks and read the usage page; it shows credits per member, per Flow and per model. For a larger rollout, talk to us and we size it from your real usage.
Why did a run cost more than the examples on this page?
Longer inputs, deeper research or more revisions. Every call carries the whole thread so far, so a run that reads five source documents and revises twice sends far more tokens than the same Flow on a short brief. The usage page shows the credits per run, and the thread shows the steps behind them.