Aether in the classroom and lab.
We give academic groups free Research-tier access, fund safety-evaluation grants, ship reproducible benchmark kits with every release, and run a civilisational-scale model series. The model that runs a customer pilot in the morning runs a graduate-thesis experiment in the afternoon.
Four programmes, free for research.
The headline programmes that sit under Apex's commitment to academic research. Each runs continuously; access is granted by application reviewed by an external panel.
Free Research-tier access
Citation-grade reproducibility, the published benchmark cases, and the smaller model variants — at zero cost for accredited academic groups. Attribution required; nothing more.
Safety-evaluation grants
Annual grants up to $250,000 to academic groups working on AI evaluation, biosecurity refusal-corpus methodology, or red-team techniques for autonomous-lab systems. Reviewed by an external panel.
Open-benchmark programme
Every Aether release ships with reproduction kits — case, mesh, solver, extraction scripts, model checkpoint hash. Academic groups can publish head-to-head comparisons under a non-commercial licence.
Civilisational-scale model series
Specialised model variants for climate, pandemic preparedness, fusion, food security — released to qualified non-commercial users at no cost. Pilot launching with three partner institutions.
Students, faculty, instructors, and the people who fund them.
Four audience cuts on Apex's academic offer. Each has a different relationship with the platform; each gets a different starting line.
- 01Open
Graduate students
Use Aether for thesis research at no cost. Reproduce a published benchmark, extend it, publish the extension. We support reasonable compute for the case you're working on.
- 02Open
Faculty research groups
Integrate Aether into your group's research. We provide model access, benchmark suites, and a forward-deployed engineer for the first project. Joint authorship welcome.
- 03Open
Course instructors
Use Aether as the platform for graduate courses in scientific ML, simulation, and AI safety. Teaching materials, reproduction kits and lecture aids on request.
- 04Open
University leadership
Discuss campus-wide access, joint research centres, or curriculum partnerships. Programmes are structured per institution — write to academic@apexworldlabs.com.
Six topics open for academic work.
The research threads where Apex has the most to offer academic groups — published benchmarks, methodology to reproduce, open questions where outside eyes change the answer.
Simulation pretraining & scaling laws
Empirical scaling exponents for simulation-native models, per-domain breakdowns, the data-quality cliff, dense vs sparse-MoE scaling. Reproduce the curves, extend them, or build alternative architectures against the same eval suite.
Cross-domain transfer
How does CFD pretraining help cell-painting? Why does discrete-state pretraining transfer poorly to continuous dynamics? Mechanism, ablations, and the conservation-law regularisation that may underlie the effect.
Long-rollout stability
Why single-step cross-entropy is only weakly predictive of long-horizon stability. Drift, error compounding, and what the curves imply for evaluating autoregressive physics models.
AI safety & evaluation
Refusal-corpus methodology, capability gating for autonomous-lab systems, biosecurity red-teaming. Safety-evaluation grants fund work in these areas; published benchmarks let external groups contribute.
Biosecurity for autonomous wet-labs
Architecture of dual-use refusals at the model layer, chain-of-custody for reagent flow, and red-team methodology for autonomous-lab systems. Published; reviewed by external biosecurity advisors.
Foundation models in regulated science
FDA AI/ML guidance, GxP-aligned model lifecycle management, audit trails as a research object. Academic groups working with regulators are first-class partners.
Sample graduate courses using Aether in the classroom.
A slice of the courses run with Apex as the platform. Teaching materials are CC-BY; instructors can adapt the syllabus and the assignments to their department's curriculum.
- ML 6.S991 (MIT)
Foundation models for the physical world
Graduate seminar on simulation-native pretraining. Uses Aether as the platform; final-project benchmarks reproduce published Apex results and extend them.
- CS 329S (Stanford)
Safety engineering for autonomous lab systems
Capability gating, refusal-corpus methodology, biosecurity evaluation. Lab portion runs on a sovereign Apex tenancy; assignments graded against the open red-team corpus.
- AERO 5301 (Cambridge)
Computational aero-elasticity with foundation models
Flutter, divergence, gust response — taught with Aether as the workflow substrate. Coursework benchmarks against measured wind-tunnel data on the published flight-test corpus.
- MAT 8702 (ETH)
Generative materials discovery
DFT screening, crystal-structure prediction, generative composition design. Final project: novel composition discovered, calibrated against measured property data.
Compute, mentorship, materials, and an open repo.
Six concrete things you get with an Apex academic engagement. None of this is hidden behind a sales conversation; published terms are in the order form.
Compute allocation
Research-tier accounts include throughput appropriate for thesis-scale work. Civilisational-scale model series users get up to 64 H100 hours per quarter; visiting fellows get dedicated compute for their engagement.
Reproduction kits
Every published Apex benchmark ships with the case definition, mesh, solver settings, extraction script and model-checkpoint hash. Pull a kit, rerun the benchmark, extend it. Citation-grade reproducibility is the goal.
Mentorship
Apex researchers pair with academic fellows during their fellowships. Joint authorship is normal; intellectual property follows the fellow. Mentorship continues after the formal engagement closes.
Teaching materials
Lecture slides, problem sets, runnable notebooks for instructors using Apex in graduate courses. Materials are CC-BY-licensed; instructors can modify, share and extend freely.
Convenings
We host an annual academic convening — invited talks, working sessions on simulation-native ML, and a public red-team exercise. Partner-university researchers attend at no cost.
Open-source
Reproduction-kit tooling, benchmark scaffolds and the open subset of the eval harness are MIT-licensed. Contributions back to the open repos are welcome; merged contributors are publicly credited.
From a one-page brief to a funded programme.
Two-business-day acknowledgement, two weeks for small grants, four weeks for material funding that goes to external review. The decision and the rationale are sent in writing; the funded brief is published.
- 01Week 0
One-page brief
Email academic@apexworldlabs.com with what you're researching, what you need (model access, compute, data, funding) and a written outcome. We respond within two business days.
- 02Weeks 1–2
Conversation
A 30-minute call with the Labs team. We assess fit; you assess whether we're useful. Either side can pull out without prejudice.
- 03Week 3
External review
Material funding decisions go to an external academic panel. Membership and tenure published; vote on grants is binding; dissents published alongside the consensus.
- 04Week 4
Decision
Acceptance, decline with reasons, or referral to a different Apex Labs programme. Decisions are sent in writing; the funded brief is published.
Eight institutions across three continents.
A slice of the academic partners running grants, fellowships or civilisational-scale pilots with Apex Labs today. Each engagement is named and reviewed; the funded briefs are public.
- MIT
- Stanford
- Oxford
- Cambridge
- Princeton
- UCL
- ETH Zürich
- UC Berkeley
Eligibility, IP, deployment, and the bits about exports.
The questions academic users send most often. If yours isn't here, write to academic@apexworldlabs.com — same business day where we can.
Who is eligible for free academic access?
Accredited academic groups, non-commercial research institutes, and 501(c)(3) / equivalent non-profits doing research that doesn't directly fund a commercial product. Industry researchers should apply via the commercial track.
What can I do with the models?
Research, publication, education, and reproduction of any published benchmark. Commercial use requires a separate licence. Re-distribution of weights is not permitted; re-distribution of derived results is encouraged.
Who owns the IP from funded work?
The funded researcher and their institution. The Apex Labs grant agreement assigns IP to the academic side explicitly. Apex retains a non-exclusive licence to reproduce published outputs for our own research and educational use; nothing more.
How does this fit into a course curriculum?
Teaching-tier access is available to instructors running graduate or advanced-undergraduate courses in scientific ML, simulation, AI safety or adjacent topics. Course materials, reproduction kits and lecture aids on request.
Can students use Apex on their own?
Yes. Individual students at accredited institutions can apply for personal Research-tier access. Approval is typically same-day for verified .edu / .ac.uk / .ac.jp / similar email addresses.
What about export-controlled academic work?
Sovereign air-gapped academic access is available under appropriate ITAR / EAR clearance. Applications reviewed quarterly. Engagement uses a cleared forward-deployed engineer.
Open, attributable, academically owned.
We want academic work on Aether to be owned by the academics doing it. Six commitments we hold ourselves to — and which are part of the charter.
Reproduction kit on every customer-facing claim
External safety advisors review programme decisions
Academic work is owned by the academic, not by Apex
Publication is encouraged; pre-publication review is optional
Grant decisions are externally reviewed and openly published
Open-source contributions credited and supported
Working on something hard?
Send us a paragraph on what you're researching. If we can help, we will. Free Research-tier access for accredited academic groups; grants reviewed by an external panel.