n8n Review (2026): 2.46 Stable, the 3.x Beta, and When to Upgrade
Published April 29, 2026 · Updated November 9, 2026 · by Jonathan Hildebrandt
The short version
n8n 2.46.7 is the current stable release, 3.1.0 is the newest 3.x beta, and stable 3.x lands in early December, after which 2.x gets 12 months of security fixes. The upgrade call by setup, and whether the 4.5 rating holds.
Pros
- ✓Agent Builder test runs execute an agent before it hits live traffic, so a broken node surfaces in a test loop instead of in a customer's inbox (n8n 2.35, August 11)
- ✓Human review is still a per-tool gate: pause the agent only on the irreversible tools and route the approval to Slack, Telegram, or n8n Chat before it fires
- ✓Local token counting shows an estimated per-run budget in the editor before you execute, so agent cost is visible up front instead of on the invoice
- ✓Self-hosting still deletes per-execution pricing, and n8n 3.0 makes Docker the only supported self-hosting path, which is where that pricing argument already pointed
- ✓n8n 3.0's security defaults got meaningfully safer: encryption key rotation is always on, Slack credentials now require a signing secret to block forged requests, and the default SSRF block list widened (n8n v3.0 breaking changes, October 27)
Cons
- ✕n8n's own docs still say it plainly: "Agents are in Preview and may change in future releases. Avoid relying on them in production workflows" (n8n MCP docs, fetched November 9, 2026), about three months after the test-run and HITL features first shipped
- ✕The 3.x line is still a beta (3.1.0, November 3), the upgrade is one way with no path back to 2.x, and stable isn't due until early December, so production instances should stay on 2.46.7 for now
- ✕Self-hosting is a Docker deployment you own, and 3.0 drops the npm and npx install paths entirely, so there's no non-container fallback left
- ✕n8n cut the AI Workflow Builder credit allowance hard since August: 50/mo on Starter and 150/mo on Pro, down from the 2,300 and 13,700 cited here in August (n8n pricing, fetched 2026-11-09)
n8n Review (2026): 2.46 Stable, the 3.x Beta, and When to Upgrade
n8n's upgrade question has a deadline now. Stable 3.x is due in early December, and from that release 2.x gets security and critical bug fixes for 12 months and no new features (Introducing n8n 3.0, October 27, 2026). The same post says the move to 3.x is one way: there's no going back to 2.x, so n8n tells you to try it on a test instance with a full backup first. Since this review's October 30 update, 2.46 became the stable line (2.46.7 as of November 9) and the beta moved to 3.1.0, published November 3 (n8n GitHub releases). The 4.5 rating holds.
The pick hasn't changed. If you can run Docker, self-host the free Community Edition and skip the execution meter entirely; 3.x makes Docker the only supported self-hosting path anyway. If you can't, the 50 EUR Pro plan is the ops-outsourcing option once you clear 2,500 executions a month (n8n pricing, fetched 2026-11-09). On the upgrade itself: Docker users with simple workflows should test 3.1.0 on a non-production instance now, anyone still on an npm install has a containerization job to do first, and a production instance with no test environment should wait for December. One platform still beats n8n for a non-technical operator, Make.
What changed since October 30
Two releases matter here, and a tag that's easy to misread. Stable moved from 2.45.4 to the 2.46 line: 2.46.4 on November 2 was the first non-prerelease 2.46, and 2.46.7 shipped November 9, while 2.45 got three more patches (2.45.5 to 2.45.7) between October 28 and 30 (GitHub releases). On the 3.x side, 3.0.1 followed the beta on October 28 and 3.1.0 landed on November 3 (GitHub releases).
| Change | Date | What it means for you | Source |
|---|---|---|---|
2.46 promoted to stable; 2.46.7 is the current stable and latest tag | Nov 2 to Nov 9 | Production pin moves from 2.45.4 to 2.46.7 | GitHub releases, Docker Hub tags |
| 3.1.0 replaces 3.0.x as the newest 3.x beta | Nov 3 | Run the Migration Report against 3.1.0 behavior, not 3.0.0 | 3.1.0 release |
| 3.1.0 reverts two 3.0.0 removals: the Execute Sub-workflow node's "Run once for each item" mode and the "Any workflow" caller policy | Nov 3 | Sub-workflow setups that broke on 3.0.0 for these reasons work again | 3.1.0 release |
2.x continues as 2.47 prereleases; Docker beta and next both point to 2.47.3 | Nov 9 | n8nio/n8n:beta gets you 2.47, not 3.x; pin the 3.1.0 tag to test 3.x | Docker Hub tags |
| Migration Report updates: per-rule resolution guidance, each workflow's project shown, notification emails open a report filtered to the recipient | Nov 3 to Nov 9 | Another reason to update to the newest 2.x before you scan | 3.1.0 release, 2.47.3 release |
Source: github.com, captured November 9, 2026.
The 2.47 line is the 2.x track running in parallel with 3.x, not a renamed beta. Both 2.47.0 and 3.1.0 were published on November 3 and compared against 2.46.0, and the 3.1.0 notes carry 2.47.0's feature list plus the 3.x-only removals, such as Edit Image versions 1 and 1.1 and a bump of the community-node API version to 3 (3.1.0 release; 2.47.0 release). Until stable 3.x ships, new features land on both lines. After it ships, features go to 3.x only (n8n blog).
Upgrade now, containerize first, or wait: the call by setup
The 12-month window is the date to plan around. If stable 3.x lands in early December as scheduled, 2.x security fixes run to roughly early December 2027 (n8n blog). Anyone still on 1.x has a tighter clock: 1.x support ends December 31, 2026, and there's no direct path from 1.x to 3.x, so the first stop is 2.x (n8n blog).
| Your setup | The call | First step | Source |
|---|---|---|---|
| Docker, mostly simple workflows | Test 3.1.0 now on a copy, move production after stable | Update to the newest 2.x, run Settings > Migration Report, back up, then start a test instance on the 3.1.0 image | n8n blog, Docker Hub tags |
n8n installed with npm or npx on a VPS | Containerize first; the version bump is the easy part | Move the same 2.x version to Docker (or the one-line setup) and confirm it runs before touching 3.x | v3.0 breaking changes |
| Production traffic, no test environment | Wait for stable in December | Use the beta window to run the Migration Report and fix or mark "won't fix" what it flags | n8n blog |
The npm row is where the real work sits. Containerizing means deciding where the database, the encryption key, and the binary-data volume live, and 3.x renames that last directory from ~/.n8n/binaryData to ~/.n8n/storage on first start (v3.0 breaking changes). Do the Docker move on 2.x, where nothing else is changing, and the 3.x upgrade becomes a tag swap.
n8n 3.0 lands in beta: what breaks, and what gets safer
n8n 3.0.0 shipped as an early-access beta on October 27, 2026, with the stable release scheduled for early December (Introducing n8n 3.0, October 27, 2026). The framing from n8n's own announcement is unusually blunt for a major version bump: this release is not new features. It is internal cleanup that "improves reliability, stability and security," clearing legacy code paths that had to wait for a breaking-change window (n8n blog, October 27, 2026). New features keep shipping on the weekly cadence regardless of version number.
Source: blog.n8n.io, captured October 30, 2026.
The practical effects split into three buckets.
Self-hosting goes Docker-only. n8n 3.0 no longer publishes a runnable package to npm, so installs run with npm or npx n8n stop working entirely; n8n ships a one-line setup script as the fast path to a Docker instance, and existing Kubernetes or Helm deployments are unaffected (v3.0 breaking changes). For the solo-dev persona this review has recommended since April, that was already the plan. For anyone still running n8n off a bare npm install, 3.0 forces the move this review already argued for.
Legacy nodes get removed, not deprecated. The Function and Function Item (legacy) nodes, the Item Lists node, the legacy OpenAI node, older Agent modes, the old Cron and Interval nodes (replaced by the Schedule Trigger), and Edit Image node versions 1 and 1.1 are gone in 3.0; workflows that still use them fail outright, and published workflows using a removed node stop activating entirely (v3.0 breaking changes). Every instance on 2.39 or later has a Migration Report under Settings, which lists the affected workflows, auto-fixes many of them as a new version you review before publishing, and flags the rest for a manual edit. Run it before you upgrade, not after.
The Edit Image rebuild is the one most workflows will actually hit. Node version 2 "uses a different image-processing library," per n8n's own changelog, and the migration carries real behavior changes: blur settings convert automatically but the Sigma parameter is gone, text placement can shift, the Get Information operation returns a normalized result structure, PNG and TIFF output now ignore the Quality setting, and a resize to zero or negative dimensions fails instead of silently succeeding (v3.0 breaking changes). If your workflows touch image blur, resize, or metadata, test the output by hand after migrating, not just the fact that the workflow still runs.
Security defaults tightened across the board: encryption key rotation is on by default, Slack credentials now require a signing secret so a forged request can't impersonate a Slack trigger, the default SSRF block list is wider, decompression limits on the Compression node are lower to block oversized "zip bomb" archives, and unverified community packages are off by default (v3.0 breaking changes). None of these change what an existing, correctly configured workflow does. They close holes a misconfigured one could fall into.
| Change in 3.0 beta | Who it affects | What to do before upgrading |
|---|---|---|
| Docker-only self-hosting | Anyone on npm or npx n8n | Move to Docker via the one-line setup or your own Compose file |
| Edit Image node v1/v1.1 removed | Any workflow with a blur, resize, or text node | Run the Migration Report, then re-test blur, resize, and PNG/TIFF output by hand |
| Legacy nodes removed (Function, Item Lists, Cron, Interval, legacy OpenAI, older Agent modes) | Older workflows built before 2025 | Run the Migration Report; most have a direct, auto-migrated replacement |
| Slack credential requires a signing secret | Anyone using Slack triggers | Add the signing secret before upgrading, or the trigger stops authenticating |
Source: n8n v3.0 breaking changes and n8n blog, October 27, 2026.
What changed since August, and what did not
The August update covered 2.35.5, which gave the Agent Builder test runs, per-tool human review, and local token counting. None of those has regressed through 2.46.7 or the 3.x beta. The MCP auto-expose toggle this review flagged as pre-release in August shipped, though n8n's docs say it is rolling out gradually from 2.36.0 and may not be visible on every instance yet (n8n MCP docs, fetched 2026-11-09). The per-tenant credential isolation flagged for multi-tenant Cloud agents still hasn't appeared in n8n's changelog or docs; treat it as unresolved.
Pricing: current euros, approximate dollars, and a credit allowance that shrank
Execution caps and per-seat pricing held flat through November 9: Starter 20 EUR/mo, Pro 50 EUR/mo, unchanged since July (n8n pricing, fetched 2026-11-09). What did not hold flat is the AI Workflow Builder credit allowance, the budget for using n8n's AI chat assistant to generate and edit workflows. This review cited 2,300 credits on Starter and up to 13,700 on Pro in August; the pricing page today lists 50 and 150.
Source: n8n.io, captured October 30, 2026.
| Plan | Price (EUR, annual, n8n pricing) | USD approx | Executions/mo | AI Workflow Builder credits/mo | Hosting |
|---|---|---|---|---|---|
| Community Edition | 0 | $0 | Unlimited | n/a | You self-host |
| Starter | 20 EUR/mo | ~$22/mo | 2,500 | 50 | n8n cloud |
| Pro | 50 EUR/mo | ~$54/mo | 10,000 | 150 | n8n cloud |
| Business | 667 EUR/mo | ~$720/mo | 40,000 | Coming soon, not included today | n8n or self |
| Enterprise | Custom | Custom | Custom | 1,000 (cloud only) | n8n or self |
Prices, execution caps, and AI Workflow Builder credit allowances per n8n's pricing page, as of 2026-11-09 (unchanged from October 30). The USD column converts at roughly 1 EUR = 1.08 USD, an approximate rate for budgeting, not billing.
Two things are worth naming directly. First, the credit cut is real and dated: whatever changed n8n's calculus between August and October, 50 credits a month on Starter covers meaningfully fewer AI-assisted workflow builds than 2,300 did, and if you were budgeting around this review's earlier figure, re-budget now. Second, Business still does not include any AI Workflow Builder credits, listed as "coming soon," which is a gap worth knowing about before you pay 667 EUR a month expecting it.
Executions remain the real wall, and the arithmetic has not changed. One workflow polling an API every fifteen minutes runs about 2,880 times a month, clearing the Starter cap by itself before any other automation fires. Pro's 10,000 absorbs most small-team loads; past that you are on Business at 667 EUR or self-hosting. A managed VPS is still the escape route: the Community Edition is a Docker image with no license fee and no execution meter, and a Cloudways droplet in the 11 EUR/mo range handles patching and backups without you also running a server by hand.
Agent Builder: production-ready yet?
Closer than in August, still not there. n8n's own documentation settles this without hedging: "Agents are in Preview and may change in future releases. Avoid relying on them in production workflows" (n8n MCP docs, fetched 2026-11-09). That line has survived every weekly release since this review first flagged the Agent Builder as Preview, including 3.1.0, even as n8n keeps adding the plumbing a production agent needs: 3.1.0 and 2.47.0 add agent version history with a publication timeline, publishing a saved agent version independently of its draft, and MCP tools that read agent session logs (3.1.0 release).
What's real inside that Preview: test runs execute an agent workflow against generated input so a broken tool node fails in the editor, not in front of a live user. Per-tool human review lets you gate only the irreversible tools. Say your agent reads a CRM, drafts a reply, and sends an email; enable review on the send-email tool only, and the agent runs the read and draft unattended, pauses before the send, and drops an approval request into Slack showing the exact recipient and body. A reviewer approves or denies, and on denial the model is told no and continues (n8n HITL docs). Token counting estimates the per-run budget before you execute, so a context-heavy run is visible before it bills, not after.
What's still missing is the reliability layer the Preview label is warning you about. The AI Agent node hands you the loop; it does not absorb malformed model output for you, and a dependable production agent still means building the retry path yourself. Turn these features on in staging. The vendor is telling you, in writing, not to bet a customer-facing SLA on them yet.
MCP integration depth
n8n's MCP server is no longer carrying the hedge its Agent Builder still does. Instance-level MCP access, the feature that lets Claude, ChatGPT, or another MCP client search, run, and build n8n workflows from a single connection, shows no Preview label in n8n's current docs, and the auto-expose toggle this review tracked as pre-release in August has shipped: enable it once at Settings > Instance-level MCP and every new workflow becomes available to connected clients without a per-workflow step. n8n notes the setting is rolling out gradually from 2.36.0, so it may not show on every instance yet (n8n MCP docs, fetched 2026-11-09). Exposing an agent, as opposed to a workflow, to an MCP client still carries the same Preview warning as the Agent Builder itself, which is the one place the hedge survives.
If you are wiring n8n into an agent stack that speaks MCP, the 2026-07-28 stateless MCP spec is the context worth reading first, because it changes how those server connections hold state. No other automation platform in this review's coverage lets a client auto-register new workflows to an MCP server with a single settings toggle. That gap is the clearest reason an agent-first team picks n8n over the alternatives.
The verdict, per persona
The rating stays 4.5. n8n's core economics (self-hosting deletes the per-execution meter, the MCP server is first-party and now out of Preview) have not moved, and the 3.x beta is a cleanup release, not a reason to upgrade the score or downgrade it. The pick, by buyer:
Solo dev, comfortable with Docker. Run the free Community Edition on a managed VPS. An 11 EUR/mo Cloudways box gives you unlimited executions and the full Agent Builder for zero license cost, and 3.x making Docker mandatory for everyone else just catches up to where this persona already was. Test 3.1.0 on a second container now; switch production once stable ships.
Ops team, 2,500-plus executions a month, no one to own uptime. Buy Pro at 50 EUR/mo (pricing), but plan around 150 AI Workflow Builder credits a month, not the 13,700 this review cited in August. If you have even one engineer who can run a container, self-host instead and put the difference into monitoring.
Non-technical operator who will not touch a container. Make still wins. The Agent Builder's test run and per-tool review gate were the friction reducers that could have closed the gap, and n8n's own docs still mark that feature Preview about three months after it shipped. Revisit this call once the Agent Builder label changes, not before.
Enterprise with multi-tenant agents. The credential-isolation fix this review has been waiting on since August has not surfaced in n8n's changelog or docs. If that is your blocker, the answer is still "not yet," and self-hosted Business or Enterprise on infrastructure you control is the safer path until n8n documents a fix.
Agent-first team that wants an MCP-native workflow. n8n is the clearest choice in the category. The MCP server's auto-expose toggle and its graduation out of Preview status are real, shipped advantages nothing else in this category matches.
What to watch
Stable 3.x, due early December per n8n's announcement, is the next trigger. It makes the Docker requirement, the node removals, and the stricter security defaults what every production instance runs, and it starts the 12-month countdown on 2.x security fixes. Watch for the Docker beta tag moving from the 2.x line to 3.x (Docker Hub tags); that is the practical signal the 3.x track has become the default. The second trigger is the Agent Builder losing its Preview label, which is what a customer-facing team needs before trusting agent workflows past staging. Until stable 3.x ships, run 2.46.7 in production, and the pick stands.
Docker is now mandatory for self-hosting, so the open question is where to run it. Stand up the self-hosted image from n8n on a box you already manage, or run it on Cloudways to get the unlimited-execution economics without patching and backing up a server yourself.
Review history
This review is updated in place as the product changes. Earlier versions:
- November 9, 2026: 2.46 Stable, the 3.x Beta, and When to Upgrade (current version)
- October 30, 2026: Is the 3.0 Beta Worth the Upgrade?
- August 20, 2026: Is the 2.35 Agent Builder Worth the Upgrade?
- July 2, 2026: 2.26 Is Out, the MCP Server Is Real, and Pricing Moved to Euros
- May 1, 2026: The Open-Source Automation Platform Worth the Setup Time
Ready to try it?
Try n8n →