Gumloop Pricing for Small Business (2026): Credits, Agents, Overage, and Real Cost

Gumloop pricing starts at $37 per month for Pro, with 20,000 monthly credits; Enterprise is custom-priced. The current offer includes unlimited seats, agents, and teams, but not unlimited usage or simultaneous work. For a small business, the important number is the cost of a completed, usable task across the shared organization—not the number of employees invited.

New accounts get a 14-day Pro trial requiring a card, which converts to paid unless cancelled. Pro credits do not roll over. Optional overage costs $0.005 per credit, and the default overage ceiling is a substantial $5,000 per billing period. Lower it before a broad rollout. BYOK removes model charges from Gumloop credits, but adds a separate provider bill and changes the agent orchestration fee.

Pricing checked September 15, 2026. This is a vendor-document-based evaluation, not hands-on testing. The official pricing page and credit documentation are the factual authorities; see the sourcing limitations below.

AI-generated explanatory diagram; not a Gumloop interface or measured usage report.

Gumloop pricing plans as of September 15, 2026

Pro is the self-serve starting point; Enterprise changes negotiated capacity and organizational controls, not simply the seat count. Dollar figures below follow Gumloop's published dollar pricing. Confirm currency, taxes and final billing terms at checkout or in your quote.

Sources: official pricing, including the same-day captured pricing-table evidence used for this review, and official credits. The credits page also identifies the Insights dashboard as an Enterprise feature.

The 20,000-credit allowance is not a promise of $100 cash value. Gumloop describes it as 7,400 credits at the $0.005 list price, plus 12,600 bonus credits. Dividing $37 by 20,000 gives a blended subscription cost of $0.00185 per included credit if you use the entire allowance. Additional credits cost $0.005 each. That distinction matters when a successful pilot expands.

Do not use an older Free/Solo/Team table as the current purchase menu. Historical third-party articles may still describe those tiers; the current documented offer is Pro and Enterprise with a one-time Pro trial. A team collaboration space is also not the same thing as an old paid “Team” tier. Existing customers should check their actual subscription rather than infer migration terms from this guide.

What a Gumloop credit actually pays for

Agent chats combine model usage, tool calls, active compute and orchestration; a credit is not a fixed task or message. The credit ledger defines one credit as $0.005 and rounds charges up to whole credits.

A $0.03 model charge therefore corresponds to six credits before the other components. One user message can trigger multiple model steps as the agent chooses tools, reads results and answers. Connected tool definitions also consume prompt tokens; adding every available connector is not free context.

The tool-call rule applies to successful calls. Failed tool calls are not charged as tool calls, but that does not establish that an entire unsuccessful conversation is free: the agent may already have used model tokens, active compute or other successful calls. Check the ledger rather than treating “failed task” as a billing category.

Waiting for a human reply is not active compute. Enterprise customers running Gumloop in their own VPC are not charged Gumloop compute, according to the docs; their own infrastructure is not thereby free.

For a simple standard-rate estimate, start with the already-accounted component credits:

Total = model credits + tool credits + compute credits + rounded-up 8% orchestration.

This is a planning model, not a replica of every billing event. The live drawer can show additional categories such as evaluation/self-improvement, subagents and model routing. Interactive artifacts that fetch live data have a separate documented runtime charge, paid by the viewer. If those features are involved, inspect those charges too; do not force every bill into a four-row estimate.

Three small-business workload examples

These are illustrative budgets, not measured Gumloop benchmarks or guarantees of output quality. Each example assumes a repeated agent task with the stated model cost, successful base-rate tool calls, whole minutes of active compute, no paid tool surcharges, and a single rounded orchestration calculation. Actual per-event rounding, retries, context, routing and other features can change usage.

The examples use separate organizations or alternative workloads—not three allowances in the same account. Each starts with one Pro subscription and its 20,000 monthly credits. Overage must be enabled with a sufficient cap for the higher-volume examples.

For the first workload, the base is 13 credits. Eight percent is 1.04, rounded up to two: 15 credits per task. For lead triage, the base is 29 and the rounded fee is three. For research, the base is 118 and the rounded fee is ten.

In either 64,000-credit scenario, billable overage is 64,000 − 20,000 = 44,000 credits. At $0.005 each, that adds $220. The lighter workload cannot bank its unused 5,000 Pro credits for next month.

The operational lesson is that 500 complex outputs can cost as much as 2,000 simpler ones. “We only have five users” is not enough to estimate a bill. Nor does paying for more credits solve a concurrency bottleneck: schedule a representative peak-load test separately.

For a real forecast, replace all assumed component costs with recorded trial usage. Count accepted outputs, rejected drafts, successful duplicate actions and human review time. A cheap first response that needs repeated correction may be an expensive completed job.

Does BYOK reduce the total bill?

BYOK reduces Gumloop model-credit consumption, but does not automatically reduce total cash spending. On Pro or higher, supported provider keys make covered model calls zero Gumloop credits; the provider bills you directly. Tool and compute charges remain.

Crucially, Gumloop documents a 16% agent orchestration fee with BYOK, calculated on what the run would have been worth before the model-credit waiver. It is not just 16% of the remaining tool and compute charges. Enterprise uses its negotiated rate plus eight percentage points.

AI-generated conceptual diagram. Compare both bills; this is not a screenshot.

Apply that rule to the third workload above. Its pre-waiver base is 118 credits:

• Standard: 100 model + 18 tools/compute + 10 fee = 128 Gumloop credits per task.
• BYOK: zero model + 18 tools/compute + ceiling(118 × 16%) = 37 Gumloop credits per task.
• Across 500 tasks, BYOK consumes 18,500 Gumloop credits, within the included allowance, but adds an assumed $250 provider bill at the same $0.50 model cost per task.
• Total assumed cash cost becomes $37 + $250 = $287, compared with $257 without BYOK.

That does not make BYOK a bad feature. It demonstrates why bonus credits, workload volume and the separate provider bill must appear in the same spreadsheet. At other volumes or provider rates, the result can reverse. Do not copy this example as a universal break-even threshold.

Each key only covers the models served by that provider. Switching providers or using uncovered models can leave Gumloop model charges in the ledger. Test the actual model selection and confirm that waived charges appear as free before extrapolating a BYOK budget.

Unlimited seats share one budget, not unlimited capacity

Gumloop bills the organization, not each team or user separately. The organization documentation says personal-space and team activity draw from the same credit pool. Unlimited seats remove a seat-multiplier from the advertised base subscription; they do not create separate allowances.

At a hypothetical ten active employees, $37 divided by ten is $3.70 each before overage. That is an allocation example, not a per-user price or guaranteed employee cost. One heavy researcher can consume more credits than everyone else combined.

Pro's 25 concurrent agent chats and five concurrent workflow runs are different limits. Unlimited saved agents do not mean unlimited simultaneous chats, and workflow capacity is not interchangeable with agent capacity. Test burst behavior and get vendor confirmation of queueing and limit scope before relying on a time-critical process.

Personal spaces are private by default. The docs' permission table says team members receive Viewer access by default and need Editor permission to edit. Shared team apps and credentials are available to team members. A later paragraph on the same page describes editing after moving an item more broadly, so verify effective permissions in your own pilot instead of assuming the documentation resolves every sharing edge case.

For governance, distinguish Pro's agent-scoped connector policies from Enterprise's organization-wide controls. The credit docs explicitly reserve Insights for Enterprise. If you need SSO, audit retention, provisioning, contractual security guarantees or a particular deployment model, make those written quote requirements; unlimited seats are not evidence that every control is included.

Cost-control checklist for the 14-day trial

Use the trial to measure one bounded process before inviting the whole business. The trial is a one-time, card-required offer and becomes a paid Pro subscription unless cancelled through Manage Subscription in Usage & Limits.

AI-generated evaluation checklist diagram, not a product workflow screenshot.

1. Set a budget owner. Remember that multiple teams consume the same pool.
2. Review overage before enabling it. The default Pro ceiling is 1,000,000 overage credits—$5,000—per billing period. Lower it to an intentional amount. For example, 10,000 overage credits corresponds to $50 beyond the subscription, excluding other bills and taxes.
3. Keep notifications useful. The default usage thresholds are 75% and 90%; the out-of-credit notification is on by default. Email alerts are not a substitute for a spending cap or operational fallback.
4. Run ordinary, difficult and exception cases. Record the model, successful calls, active time, total credits and whether the result was usable. Include repeated runs rather than one polished demonstration.
5. Inspect the charge details. The chat header opens a breakdown. Credit logs provide grouped and detailed views; organization admins/managers can inspect usage across users.
6. Reduce avoidable context. Connect only needed tools, start fresh conversations for unrelated jobs, and select an adequate model rather than automatically using the most expensive one.
7. Reconcile BYOK separately. Add provider invoices and coverage gaps to Gumloop spending. Do not mistake a zero model-credit line for zero cost.
8. Plan for a stop. Agents stop when the overage cap is reached. Keep a manual route for work that cannot wait and confirm behavior when the included allowance runs out with overage disabled.
9. Choose deliberately before trial expiry. Keep, cancel or request Enterprise terms based on observed quality, capacity and governance—not the number of agents created.

When agents fit—and when deterministic workflows fit better

Use agents when interpreting variable inputs is the job; prefer deterministic logic when the correct steps are already known. Summarizing differently formatted requests, researching a question and drafting an internal answer can justify model reasoning. Copying a known field from one record into another usually needs less discretion.

An agent-led lead-triage process might interpret notes and suggest a category, while fixed rules validate required fields and a human approves an external action. Keeping the uncertain judgment separate from predictable writes makes failures easier to investigate. It does not eliminate setup work, permissions, rate limits or monitoring.

Gumloop's pricing presentation separates current agents from legacy workflows. Do not apply the agent-chat ledger above as a universal quote for every workflow node. Verify flow-specific consumption from the actual run and applicable documentation before budgeting a migration.

For broader comparisons, see our Make AI Agents pricing guide, Zapier pricing guide and n8n alternatives guide. Compare the cost of the same completed business outcome. Gumloop credits are not Zapier tasks, Make credits or n8n executions; there is no honest one-for-one conversion.

Hidden costs and scenario verdicts

The subscription is only one part of operating cost. Budget for paid enrichment or scraping, separate provider bills under BYOK, integration access, onboarding, human review and maintaining ownership when staff change. A successful tool call can still write an unwanted duplicate; billing success is not business success.

The next step is not to buy more credits pre-emptively. Run a bounded evaluation through the official Pro trial, then price the measured workload. Ask for Enterprise when a demonstrated governance or capacity requirement—not headcount alone—makes it necessary.

Methodology and source limitations

This guide uses official documentation and illustrative arithmetic, not hands-on testing. On September 15, 2026, we directly retrieved Gumloop's credit documentation and organization/team documentation. The pricing table uses a supplied same-day verification record of the official pricing page; our own text retrieval of that page exposed navigation rather than its rendered plan table. Verify checkout and Enterprise quote terms before purchase.

The changelog retrieval likewise returned a navigation stub, so we do not assert an exact date for the shift to the current agent-oriented offer. Search-result snippets were used only to identify buyer questions and conflicting older plan descriptions, never as pricing authority. No first-party site-performance data, customer usage measurements or ROI claims informed the verdicts. Generated diagrams explain concepts; none is a product screenshot or evidence of a tested deployment.