Skip to content
Apex
Newsletter · the Apex monthly

One email a month. The good parts.

Research progress, a named customer win, model-card diffs, an open question we'd like to hear from you about, and three things we read. Sent on the first of the month — never more, never less. Reply-to a real engineer.

Research log

A short, infrequent email when we publish something new. Unsubscribe in one click.

First of the month
exactly · no exceptions
5 sections
same shape every issue
2 min
average read time
Reply-to
a real engineer reads it
Anatomy of an issue

The shape, every month.

Each issue has the same five sections in the same order. The newsletter is short, written like a memo, and engineered to be skimmable in two minutes — with links out for anyone who wants the full version.

  • Section 01

    Research progress

    What shipped, what's about to ship, what we're stuck on. Links to papers, model cards and benchmark updates.

    Sample · May 2026

    May · Aether v1.0 GA across every discipline. The 280B sparse-MoE checkpoint is in production. We published the architecture diagram in full at /simos. Loss curves and ablation tables in the paper at /research/simos-1-0.

  • Section 02

    Customer wins

    One named customer story per month, with the metric, the comparison and a link to the case study.

    Sample · May 2026

    May · A named aerospace prime moved a structural-margin workload from the incumbent. 8 weeks pilot. Result: 6.4× cycle-time reduction at iso-fidelity. Comparison memo co-authored. Read the full story at /customers/aerospace-prime-cae.

  • Section 03

    Model-card updates

    Behaviour changes, refusal additions, deprecations. The diff vs last release, in plain language.

    Sample · May 2026

    May · Refusal corpus extended with two CBRN-adjacent categories. False-positive rate on benign chemistry queries held at <0.2%. Edge 1.3B graduated from preview to GA. Full diff at /simos#card.

  • Section 04

    An open question

    Something we're still working out — and an invitation to send us your view. Some of these have shaped the roadmap.

    Sample · May 2026

    May · What is the right eval cadence for a workload that takes weeks to produce ground truth? Reply to this email with your view. We will publish a synthesis in the August issue.

  • Section 05

    What we read

    Three to five papers, posts or talks we read this month. Not exhaustive — just the ones that moved us.

    Sample · May 2026

    May · (1) On compute-bound vs IO-bound forward-rolling, Tanaka et al. (2) The case for sovereign foundation models, Holst & Mercier. (3) Why public benchmarks plateau, Bauer.

Latest issue · in full

May 2026 — read it right here.

The current issue. No signup gate; we publish every issue on the web the same day it ships to the list.

The Apex monthly · vol. 12 · 2026-05-01

v1.0 ships, the architecture diagram, and a question on long-horizon eval

From the desk of Apex research and engineering. Two-minute read; links to the underlying papers, model cards and case studies inline.

01 · Research progress

Aether v1.0 is now generally available across every discipline surface. The 280B sparse-MoE checkpoint is in production. We published the architecture diagram in full at /model — including the modality fan-in and the discipline-head decoder routing. Loss curves and ablation tables in the paper at /research.

02 · Customer wins

A named aerospace prime moved a structural-margin workload from the incumbent. Eight-week pilot. Result: 6.4× cycle-time reduction at iso-fidelity vs the legacy CAE solution. Comparison memo co-authored with their CAE director. Story at /customers.

03 · Model-card diffs

Refusal corpus extended with two CBRN-adjacent categories under the tool-policy layer. False-positive rate on benign chemistry queries held at <0.2% on the internal red-team set. Edge 1.3B graduated from preview to GA. Full diff at /model#card.

04 · An open question

What is the right eval cadence for a workload that takes weeks to produce ground truth? Reply to this email with your view. We will publish a synthesis in the August issue.

05 · What we read
  • (1) On compute-bound vs IO-bound forward-rolling — Tanaka et al.
  • (2) The case for sovereign foundation models — Holst & Mercier.
  • (3) Why public benchmarks plateau — Bauer.
Reply-to: research@apexworldlabs.com — a real engineer reads what comes back. Unsubscribe in one click. Forward freely.
What we cover

Six topics, woven into every issue.

The list is not segmented — there is one Apex monthly, and the same shape every issue. But the topics below recur, and the archive is tagged so you can read by topic on the web.

  • Research

    Papers, model-card updates, scaling-law results, internal seminars worth sharing.

  • Engineering

    Solver-tool changes, agent-runtime updates, API surface evolution, deprecations.

  • Customers

    Named customer stories, comparison memos, pilot retrospectives.

  • Safety + trust

    Refusal corpus diffs, red-team findings, compliance milestones, incident post-mortems.

  • Hiring + culture

    Open roles, team additions, internal events, what we're learning.

  • Strategy

    Founder memos, year-ahead notes, why-we-built-this decisions.

From a reader

“The Apex monthly is the only AI-lab newsletter I open the same day it lands. Two minutes, every section pays rent.”

Archive

Past issues, readable in full.

Every issue stays on the site after it ships. No paywall, no signup required, no auto-deleted backlog. Tagged so you can read by topic.

Read it however you read

Email, RSS, web. Your call.

We meet you where you read. Email is canonical; RSS is full-text; the web archive is permanent.

From readers

What people say about the monthly.

Unsolicited replies to the list. Names withheld; titles and contexts are real and approved.

  • “The only AI-lab newsletter I open the same day it lands. Two minutes, every section pays rent.”
    — Director of CAE, aerospace prime
  • “Open-question section has changed how we think about active learning. Replied once; they replied back the next day.”
    — Research lead, biotech
  • “I forward it to my whole team. Best signal-to-noise of any frontier-lab dispatch I subscribe to.”
    — Head of platform engineering, fabless semi
Who writes it

Three desks, one voice.

The newsletter is written by the people doing the work, not a comms team describing it. Editors rotate so the taste doesn't get stale.

  • A. Singh
    Founder & CEO

    Writes the founder memo when there's one. Replies personally to the most interesting open-question replies.

  • Research desk
    Apex research team

    Runs the research-progress and what-we-read sections. Rotates monthly so the taste doesn't get stale.

  • Customer desk
    Forward-deployed engineers

    Sources the customer-win section. Co-authors the story with the customer practitioner.

FAQ

The things people ask first.

  • How often, really?

    Once a month. Always the first of the month. If we miss the first, we skip the month — we don't slide it.

  • What if I want only research?

    We don't fragment the list. Five sections, same order, every issue. Skim what you don't care about; it takes ten seconds.

  • Do you sell my email?

    No. The list is operated under our privacy notice at /legal/privacy. We don't share with marketing partners, period.

  • Can I reply?

    Yes. Reply-to is a real Apex engineer; we read every reply. Some replies have ended up in the next issue's open-question section.

  • Will you cross-post to my company's internal list?

    Yes, gladly. Email newsletter@apexworldlabs.com and we'll set up a verified relay so internal forwarding stays on-brand.

  • Is there an RSS feed?

    Yes, at /newsletter/rss.xml. The same content as email, in your reader.

  • Do you have a newsletter for academics?

    Same list. Research section is denser, and our /labs programme has its own occasional updates for partner institutions.

  • Why a newsletter, not a blog?

    A blog rewards constant output; a monthly newsletter rewards consequential output. We think the second matches frontier-lab work better.

Newsletter principles

What we promise about this list.

If the newsletter breaks one of these, write to newsletter@apexworldlabs.com and tell us. We will fix it before the next issue ships.

  • 01Sent on the first of the month, never more.
  • 02Five sections, in the same order, every issue.
  • 03No marketing-speak, no 'we're excited to announce.'
  • 04Links to the underlying paper, code or model card — always.
  • 05Reply-to is a real engineer. We read what comes back.
  • 06Unsubscribe in one click. We won't try to win you back.
Subscribe

One email a month. The first of the month. That's it.

No drip campaigns. No re-targeting. No surprise emails. Reply-to a real engineer; unsubscribe in one click.

Research log

A short, infrequent email when we publish something new. Unsubscribe in one click.