Make Review August 2026: Maia Builds Your Scenarios, Gemini 3.7 Flash Lands, and Whether the 4.5 Holds

Make dropped a natural-language co-worker called Maia into the Scenario Builder on August 12, and on August 21 it added Google's Gemini 3.7 Flash to its AI modules (help.make.com/2026, retrieved 2026-08-21). Maia builds, edits, and troubleshoots scenarios from plain-text prompts. Gemini 3.7 Flash is a cheaper, faster model choice for the agents you already run. The rating holds at 4.5, the same number our July review gave it, and this piece is about why two real additions do not move that score.

Here is the one thing to leave with. Maia genuinely speeds up building and iterating, but it cannot read the execution log from a run that already failed, and it cannot see any scenario other than the one open in front of it (help.make.com/introduction-to-maia-by-make, retrieved 2026-08-21). For the ops lead running ten or more production scenarios, that is precisely the wall you hit when something breaks after hours. Maia handles the greenfield work and stops short of the debugging that actually costs you sleep.

What changed since July 17

The July review is 35 days old and predates all of this. Eight dated items, the two that carry the release first.

• Maia, August 12. An AI co-worker in the Scenario Builder that creates, modifies, and troubleshoots scenarios from prompts, in public beta on all paid plans (help.make.com/2026, retrieved 2026-08-21).
• Gemini 3.7 Flash, August 21. Google's newest Flash model is now selectable in Make's AI modules; the same release removed one prior OpenAI module (help.make.com/2026, retrieved 2026-08-21).
• Cycles-per-run removal announced, August 13. Make will remove the Cycles per run setting on October 1, 2026, which changes how looping scenarios meter credits (help.make.com/2026, retrieved 2026-08-21).
• Custom roles, August 5. Enterprise admins can now create custom organization and team roles for precise access control (help.make.com/2026, retrieved 2026-08-21).
• Opus 5 and Gemini 3.6 Flash, July 31. Both models added, alongside lower credit costs for OpenAI models (help.make.com/2026, retrieved 2026-08-21).
• Dark mode, July 30. Light, dark, and system themes (help.make.com/2026, retrieved 2026-08-21).
• Sub-agents and tool output filtering, July 27. Both landed in the Make AI Agent (New) app (help.make.com/2026, retrieved 2026-08-21).
• AI error explanations, July 24. Make AI now explains why a run failed, shipped with a scenario trash and module renaming (help.make.com/2026, retrieved 2026-08-21).

Maia and Gemini 3.7 Flash are the reason anyone reruns this review. Custom roles matter to Enterprise readers. The cycles change is a cost alert I unpack below.

Maia: what it does and where it stops

Maia answers the most common beginner complaint about Make, which is "I know what I want but not which modules to wire." Click Create scenario, describe the workflow in the chat panel on the left of the builder, and Maia assembles the modules and AI agents for you (help.make.com/introduction-to-maia-by-make, retrieved 2026-08-21). Tell it to add a Slack notification or swap the summarizer from OpenAI to Claude, and it edits module settings in place. After you click Run once, it can read that run's execution log and explain or fix the error it produced.

Availability is generous for a beta. Paid plans get unlimited access with the first 70 messages free for 30 days after your first interaction, then each reply draws credits (help.make.com/introduction-to-maia-by-make, retrieved 2026-08-21). Free-plan users get a 30-day trial with 45 messages. You work in a single chat session per scenario, Make stores those sessions for 15 days, and you can revert to any earlier version Maia produced.

The cost is where the beta label earns its keep. Each Maia reply consumes a variable number of credits based on task complexity and the tokens its model burns, and those credits come from the same pool your scenarios run on (help.make.com/introduction-to-maia-by-make, retrieved 2026-08-21). Make's own screenshots show single messages costing 140.04 and 160.18 credits (help.make.com/introduction-to-maia-by-make, retrieved 2026-08-21). On a Core plan that buys 10,000 credits for 9 USD a month (make.com/en/pricing, retrieved 2026-08-21), a chatty build session where Maia rebuilds a scenario a dozen times can quietly eat a meaningful slice of the month.

Now the ceiling, and it is a hard one. Maia doesn't have these capabilities yet (help.make.com/introduction-to-maia-by-make, retrieved 2026-08-21):

• It can't view execution logs from past runs, so it can't analyze an error after the fact. You have to describe the failure to it yourself.
• It can't see other scenarios, which rules out starting from an existing scenario, referencing another scenario's module configuration, or connecting to subscenarios and webhooks.

Read those two limits together and the shape of Maia comes into focus. It is a builder's assistant, not an operator's. A first-time user describing a three-app workflow gets a working draft in minutes. The ops lead whose 40-module scenario fails at row 380 of an iterator cannot hand Maia the log and say "find it," because Maia never sees that log. You still open the run history and read the bundle yourself, then paste the error text into the chat for a fix. That is faster than nothing, and slower than the "AI debugs your production stack" pitch the category keeps making. Maia moves the build-time needle and leaves the run-time needle roughly where it was.

Gemini 3.7 Flash is now in Make

Google shipped Gemini 3.7 Flash on August 13 as its "most intelligent workhorse model yet for coding and agents," and Make added it to the AI modules on August 21 (VentureBeat, August 13 2026; help.make.com/2026, retrieved 2026-08-21). The pitch is cost. Through December 31, 2026, it runs at an introductory 0.75 USD per million input tokens and 3.75 USD output, rising to 1.50 / 7.50 on January 1, 2027, against Claude Sonnet 5 at 2.00 / 10.00 in Google's own comparison table (VentureBeat, August 13 2026).

The benchmark case is real but lopsided, and the caveat matters: Google ran these numbers itself, and independent replication is not yet published (VentureBeat, August 13 2026).

Benchmark (Google's table)	Gemini 3.7 Flash	Claude Sonnet 5	Source
AutomationBench (workflow automation)	30.4%	10.7%	VentureBeat, Aug 13 2026
GDP.PDF (complex PDF comprehension)	34.0%	28.0%	VentureBeat, Aug 13 2026
FrontierCode 1.1 Main (production code)	43.6%	42.7%	VentureBeat, Aug 13 2026
Agent's Last Exam (desktop / OS tasks)	26.3%	33.3%	VentureBeat, Aug 13 2026

The AutomationBench gap looks decisive and is the one number to distrust most, because it is a Google benchmark measuring Google's stated strength on a Google-run test. Claude still leads Agent's Last Exam (VentureBeat, August 13 2026), and third-party trackers put the two within a point on general intelligence, with Artificial Analysis scoring Gemini 3.7 Flash at 56 against Sonnet 5 at 55 (felloai, August 13 2026). The recommendation for anyone already running Claude Sonnet 5 in an agent: clone one production scenario, point the copy at Gemini 3.7 Flash, and compare output quality and credit draw on your own workload before you switch the fleet. The intro price makes the experiment cheap; the AutomationBench headline is not evidence enough on its own.

October 1 credit change: who it affects

On August 13, Make announced it will remove the Cycles per run setting on October 1, 2026 (help.make.com/2026, retrieved 2026-08-21). This is a quiet change with a loud potential bill. Cycles per run governed how a scenario grouped its processing cycles inside a single execution, and scenarios that relied on it to batch work will execute differently once it is gone.

Who should care: anyone running high-fan-out loops, iterators over large arrays, or retry logic that reprocesses records. In Make, every module operation and every agent tool call is a credit, so a change to how cycles are grouped can shift a scenario from one metered pattern to another. The exposure is not a fixed dollar figure Make published, so treat this as an audit item rather than a number. Before October 1, open your looping and retry scenarios, note their current credit-per-run baseline from the run history, and rerun them after the change to confirm the meter reads the same. If a scenario that costs 200 credits a run suddenly costs 600, you want to catch that in a test, not on the invoice.

Make vs n8n, August 2026

For a mixed-technical team that wants managed per-member spend caps and custom access roles without standing up a server, Make edges n8n; for a fully technical team that can run and patch Docker at volume, n8n's self-hosted model still deletes the per-credit meter and wins on cost, as our n8n review lays out. Maia and Gemini 3.7 Flash do not change that split, because n8n also points at your own model keys and self-hosting is the axis Make still does not compete on.

Pricing, retrieved today

Make's plans are unchanged from July. Every price below is per month for the base 10,000-credit tier; the pricing page slider raises credit volume at adjusted rates (make.com/en/pricing, retrieved 2026-08-21).

Plan	Price / month	Credits / month	Source
Free	0 USD	1,000	make.com/en/pricing, 2026-08-21
Core	9 USD	10,000	make.com/en/pricing, 2026-08-21
Pro	16 USD	10,000	make.com/en/pricing, 2026-08-21
Teams	29 USD	10,000	make.com/en/pricing, 2026-08-21
Enterprise	Custom	Custom	make.com/en/pricing, 2026-08-21
Try Make	Start on the free plan

The verdict, by who you are

Make earns 4.5 in August 2026. Maia is a real building aid and Gemini 3.7 Flash adds a cheaper model, but Maia's no-log-access limit caps the debugging work an operator cares about most, and the two structural cons are untouched. The picks, by situation:

• Solo non-developer operator. Make, still, and Maia makes the case stronger for you specifically. You are usually building fresh scenarios, which is exactly what Maia is good at, and the first 70 messages are free for 30 days (help.make.com/introduction-to-maia-by-make, retrieved 2026-08-21). Start on Free's 1,000 credits, describe a real workflow to Maia, then move to Core at 9 USD a month once it works (make.com/en/pricing, retrieved 2026-08-21). If you land here, Make's paid plans are the buy.
• Mixed-technical ops team. Make edges n8n, and August widens the reason. Private Spaces already gave you per-member credit caps on every plan, and custom roles now let an Enterprise admin define precise organization and team permissions (help.make.com/2026, retrieved 2026-08-21). Teams runs 29 USD a month for 10,000 credits with no server to patch (make.com/en/pricing, retrieved 2026-08-21). The flip condition holds: if every member is technical and the team wants zero per-run cost more than managed governance, self-hosted n8n wins.
• Dev team with Docker capacity. n8n still wins on cost, and nothing in August moves that pick. Make added no self-hosting, so the per-credit meter stays, and Maia can't reference your existing scenarios anyway. Gemini 3.7 Flash is available on both, since both point at your own model keys. Once monthly operations clear roughly 30 to 40k and someone can run a Docker box, self-hosting removes the meter Make charges on. The line that math never shows is ops time; if nobody wants the pager, Make's managed model is still the rational buy even where the dollar figure favors n8n.

FAQ

Did Make's rating change in August? No, it holds at 4.5. Maia adds a real building aid on every paid plan and Gemini 3.7 Flash adds a cheaper model, but Maia can't read past execution logs or reference other scenarios (help.make.com/introduction-to-maia-by-make, retrieved 2026-08-21), and no self-hosting plus credit fan-out at volume keep it under a 4.7.

What is Maia and what does it cost? Maia is Make's AI co-worker in the Scenario Builder that creates, modifies, and troubleshoots scenarios from prompts, in public beta on all paid plans. Paid users get the first 70 messages free for 30 days, then each reply draws credits from the same pool as scenarios; Make's docs show single messages at 140 to 160 credits (help.make.com/introduction-to-maia-by-make, retrieved 2026-08-21). Free plans get a 30-day trial with 45 messages.

Should I switch my Make agents to Gemini 3.7 Flash? Test first. It is cheaper at an intro 0.75 / 3.75 USD per million tokens through December 31, 2026, versus Claude Sonnet 5 at 2.00 / 10.00, and it leads Google's own automation benchmark 30.4% to 10.7% (VentureBeat, August 13 2026). Google ran that benchmark itself and Claude still leads Agent's Last Exam, so clone a production scenario, point the copy at Gemini 3.7 Flash, and compare quality and credit draw on your workload before switching.

What is the October 1 cycles change? Make will remove the Cycles per run setting on October 1, 2026 (help.make.com/2026, retrieved 2026-08-21). It changes how scenarios group their processing cycles per execution, which can shift credit usage on looping and retry scenarios. Audit those scenarios and compare their credit-per-run before and after the change.

What is current Make pricing? Free is 1,000 credits a month. Core is 9 USD, Pro 16 USD, Teams 29 USD, each for 10,000 credits a month, with Enterprise custom (make.com/en/pricing, retrieved 2026-08-21).