Oplo Center for Research & Discovery
OCRD is Oplo's open research center. We study intelligence, computation, and their consequences — from the transistor to the society — and publish what we learn.
Our work
We do research in the open, and share it with the world.
Papers, code, and datasets released for anyone to read, reproduce, and build on. Progress compounds when it's shared.
We hold our results to the standards of the fields we work in, and report what we find — including when it surprises us.
Every advance is measured by its effect on real people. Safety, interpretability, and society sit at the center, not the edge.
Research areas
OCRD spans the full stack of modern discovery — the silicon that computes, the models that reason, and the people and institutions they touch.
Training, reasoning, and multimodality at scale — and the science of making frontier models more capable with less.
142 publications → 109Reading the internals of neural networks and steering them — so that powerful systems stay understandable and safe.
88 publications → 100How people and intelligent systems collaborate — interfaces, trust, and the design of tools that augment human judgment.
63 publications → 10−9The chips, compilers, and distributed systems that make discovery run — from custom accelerators to planetary-scale training.
97 publications → 106Machine learning for materials, weather, biology, and energy — accelerating the sciences that address the physical world.
54 publications → 108Measuring technology's effect on people, work, and institutions — nonpartisan data on how intelligent systems reshape the world.
61 publications →Publications
A selection of peer-reviewed papers, technical reports, and open datasets released by OCRD teams.
We derive empirical scaling relationships that predict how reasoning accuracy improves with compute, data, and modality mix — and show that carefully balanced multimodal training reaches parity with text-only models using 40% fewer FLOPs.
A. Verma, L. Okafor, R. Tanaka, and 9 others · Featured · June 2026
M. Haddad, S. Nolan, P. Rao
J. Fields, E. Moreau, and the Science team
D. Alvarez, K. Osei, T. Lindqvist
N. Bianchi, F. Adeyemi, and the Society team
C. Wren, H. Park, O. Salcedo
By the numbers
OCRD's output is public by default. Everything below is free to read, download, and build on.
Featured finding
Tracking a fixed suite of reasoning, coding, and science benchmarks since 2021, aggregate capability rises about an order of magnitude annually — a straight line on a log scale. Understanding what bends that line, up or down, is central to OCRD's work.
Source: OCRD Capability Index, 2021–2026 · n = 40 benchmarks
OCRD Capability Index — log scale
Newsroom
OCRD brings together researchers, engineers, and scientists who want their work to be read, reproduced, and put to use. If that's you, we'd like to meet.