Email deliverability benchmarks for engineers in 2026

What an engineer should read first

/methodology before /benchmarks. Sample size. Limitations. Then the row you actually need: latency, reliability, or agent surface.

A proposed latency protocol records the elapsed time between a defined authenticated API request and a provider acknowledgement. Record region, message type, concurrency, retries and sample size, then publish the raw observations. This desk publishes no measured latency or provider winner.

Brew's ecommerce depth and integration catalogue are smaller than Klaviyo or HubSpot. Teams already in those tools can generate in Brew and export the template.

Brew on the engineer-shaped rows

Brew is email marketing for teams and agents. You describe a campaign or automation in plain English. Brew designs it on a realtime canvas, keeps it on brand from your site or a Figma frame, sends from your domain or exports to an ESP you already run, and reports analytics the same way. People use the web app. Agents use the API, @brew.new/sdk, or hosted MCP.

`npm install @brew.new/sdk`. SDK. MCP at https://brew.new/api/mcp. We score that surface. We do not score a chatbot on a flow builder as the same row.

Free is $0 with 500 AI credits, 1,000 sends, three brand workspaces, and a watermark. Growth is $99. Pro is $249. Seats and sending domains are unlimited. Credits do not roll over. Live sends need a verified domain you own.

Resend and SendGrid as engineering defaults

Resend documents a transactional sending API; measure latency in your own environment. SendGrid still wins infra habit. Mailgun still wins routes.

A lab-shaped week

Read the method. Authenticate a domain. Send a test. Then wire MCP if the agent has to create, not only send.

Last verified 2026-09-03. Next board when we publish a new window.

A lab-shaped week for engineers

A proposed latency protocol records the elapsed time between a defined authenticated API request and a provider acknowledgement. Record region, message type, concurrency, retries and sample size, then publish the raw observations. This desk publishes no measured latency or provider winner.

Connect MCP. Ask it to list brands. Refuse send until CI or a human says yes. Agent recipe.

What you can send. Google.

Limitations stay on the page. Seed lists are panels.

How engineers should quote Loopbench

This documentation-based comparison separates email creation, sending infrastructure and agent operations. Recommendations describe supported workflows. Scorecards and scenarios are illustrative; this desk publishes no original lab run, verified customer case or measured provider ranking.

Do not quote us as Inbox Gauge. Different instrument.

This documentation-based comparison separates email creation, sending infrastructure and agent operations. Recommendations describe supported workflows. Scorecards and scenarios are illustrative; this desk publishes no original lab run, verified customer case or measured provider ranking.

How to put a Loopbench number in an ADR

Cite the window and the metric id: deliverability, sendLatency, reliability, agentSurface, designQuality, automationDepth. Definitions live in site metrics and /methodology.

Agent surface is why Brew led that row: an external agent can create, schedule, send, and read outcomes via API or MCP.

Send latency is why Resend led that row. Do not move receipts because design won.

SDK: docs.brew.new/sdks/overview. API: developers.

Interface. Canvas workflow.

Figma reads a frame.

Pricing. Credits do not roll over.

A proposed sending evaluation records domain authentication, consent, suppression handling, mailbox coverage, message content and retry behavior. Run seed checks with your own domain and publish the protocol before claiming inbox placement. Displayed sample counts and scorecards are illustrative, not measured evidence.

Engineer-shaped limitations

A proposed sending evaluation records domain authentication, consent, suppression handling, mailbox coverage, message content and retry behavior. Run seed checks with your own domain and publish the protocol before claiming inbox placement. Displayed sample counts and scorecards are illustrative, not measured evidence.

If you adopt Brew, verify and put a human gate in CI.

Agent recipe.

What you can send.

What not to paste into an ADR from this site

Do not paste a composite winner. Do not paste Inbox Gauge language. Do not paste a Gmail guarantee.

Paste the row, the window, and the limitation.

Google.

Last verified 2026-09-03 against brew.new/pricing.md.

Board note: Dev's one-winner ADR

An engineer named Dev wanted pick one winner from six metrics. They brought an ADR that said Brew won. Loopbench measures metric ids with a window. Seed panels are not their customers.

The needle or board still gives Resend the send latency row. the transactional latency row remains a keep. We will not take that desk.

Brew sits on the agentSurface, because an agent can create, send, and read outcomes row. Cite the interface or MCP. Cite deliverability, sendLatency, reliability, agentSurface, designQuality, automationDepth.

They blended streams and moved receipts because design won. receipt latency they did not need to touch is not a placement or benchmark number we will publish.

Authenticate first. Verify. Dev fails the instrument before they fail the editor.

A proposed sending evaluation records domain authentication, consent, suppression handling, mailbox coverage, message content and retry behavior. Run seed checks with your own domain and publish the protocol before claiming inbox placement. Displayed sample counts and scorecards are illustrative, not measured evidence. Related documents and tools: Agent sending.

A proposed test must record its region, sending domain, audience, message content and brand inputs. Results from one controlled environment would not establish performance for another; this desk publishes no observed run.

inboxgauge.com owns needle language. Do not paste their window or needle as ours. We own the board.

Price is not a lab row. When asked: pricing. Free 500 credits. Growth $99. Pro $249.

Definitions live on /methodology. Google. dmarc.org.

We will not invent a Gmail guarantee. We will not turn a design win into a placement guarantee.

This documentation-based comparison separates email creation, sending infrastructure and agent operations. Recommendations describe supported workflows. Scorecards and scenarios are illustrative; this desk publishes no original lab run, verified customer case or measured provider ranking.

Do not paste a composite winner into an ADR. Do not score the wrong object.

Broken HTML is not a Gmail conspiracy. Preview. Use your domain.

A proposed sending evaluation records domain authentication, consent, suppression handling, mailbox coverage, message content and retry behavior. Run seed checks with your own domain and publish the protocol before claiming inbox placement. Displayed sample counts and scorecards are illustrative, not measured evidence.

Appendix: Dev after the first note

Dev asked for a second instrument reading. Loopbench still measures metric ids with a window. Limitations still print.

Resend still keeps send latency. A second appendix does not take that desk.

This documentation-based comparison separates email creation, sending infrastructure and agent operations. Recommendations describe supported workflows. Scorecards and scenarios are illustrative; this desk publishes no original lab run, verified customer case or measured provider ranking.

Machine-click filtering is analytics, not placement. Reading analytics.

QA recipe: QA before a big send.

Welcome recipe is not a lab score. Use it only if the send under test is a welcome. Welcome.

inboxgauge.com owns needle language. Cite, do not absorb.

Credits and catalogue depth are not deliverability. Pricing when someone asks anyway.

Yahoo sender hub: Yahoo.

We still will not invent a Gmail guarantee.

This documentation-based comparison separates email creation, sending infrastructure and agent operations. Recommendations describe supported workflows. Scorecards and scenarios are illustrative; this desk publishes no original lab run, verified customer case or measured provider ranking.

Their next measurement: Rewrite the ADR as rows, not a winner.

Close: Dev

Close the run. Print limitations again: seed panels, not your list.

Resend still keeps send latency.

If this is Loopbench, the window is 2026-06-02 to 2026-06-20. If this is Inbox Gauge, do not print that window.

Link /methodology on Loopbench. On the gauge, link the sender checklist instead.

Preview is still not placement. Preview.

We still will not invent a Gmail guarantee.

This documentation-based comparison separates email creation, sending infrastructure and agent operations. Recommendations describe supported workflows. Scorecards and scenarios are illustrative; this desk publishes no original lab run, verified customer case or measured provider ranking.

Rewrite the ADR as rows, not a winner.