Open Publication for Universal Science
OPUS
A journal worthy of the science it carries.
Every paper is peer-reviewed with rigor and published as a structured, citable object — readable by humans and agents alike, and carried from publication toward real-world impact.
c-0001 · result · △ verified · cited ×3
Agentic systems recover 87.4% of ground-truth claims.
- Open
- No fee to publish or to read.
- Typed
- Every claim is an addressable object.
- Reviewed
- Rigorous review, in hours.
- Dual-read
- First-class for humans and agents.
Scientific publishing is overdue for a rebuild.
Trace a finished result down the pipeline and it snags — five times. Each friction is a place where knowledge slows, leaks value, or stops moving altogether.
The pipeline from a finished result to real-world value, with its five points of failure marked. Select a clamp to inspect it.
Papers are PDFs.
The format is opaque to agents and imprecise even for humans — hard to cite exactly, hard to verify claim by claim.
0 machine-addressable claims
Every paper is a typed object, not a flat document.
Every paper is a typed object — not a PDF.
A claim, its evidence, and its citations are stored as data. The same record renders as a paper for a human and resolves as an API for an agent. Flip between the two — it is one object the whole time.
A reproducible benchmark for agentic literature synthesis across the life sciences.
L. Bertrand · M. Okafor · S. Hassan · J. Müller · A. Collet
We introduce OPUS-Bench, a benchmark for literature synthesis by autonomous agents. Agentic systems with OPUS API access recover 87.4% of ground-truth claimsc-0001, while retrieval over raw PDFs reaches only 62.1%c-0002. Across all tasks, structured access cuts synthesis error by 41%c-0003, suggesting structure — not scale — is the binding constraint.
- id
- c-0001
- type
- result
- evidence
- §4.2 · fig.3
- builds_on
- c-0002
- cited_by
- 3
- reproduction
- △ verified
Touch a highlighted claim in the abstract — it is the same object on both sides of the glass.
Reference and build on any single claim.
A citation on OPUS resolves to a claim, not a document. Because the record is built from typed claims, scientific lineage is exact — you can see precisely what a result rests on, and what now rests on it.
Hours, not months — and always credited.
Review on OPUS happens in a focused, structured tool. AI triage clears the mechanical checks; humans judge the science. The work is signed, timed, and counts toward the reviewer's standing.
Both tracks carry a manuscript from submission to decision. The OPUS bar is enlarged roughly 14× for legibility — the real speed-up is closer to 90×.
- ◉R. Mensahstructural biology14 reviewscredited 6×
- ◉A. VossML systems9 reviewscredited 4×
- ◉P. Aliyevmaterials science11 reviewscredited 5×
Every review is a citable contribution on the reviewer's record — no longer invisible, unpaid labour.
Publishing should be a beginning, not an end.
A published result is often the seed of something larger — a tool, a method, a venture. OPUS does not stop at publication; it gives a finished paper a real path toward real-world value.
A finished result enters the record as a canonical, typed object — peer-reviewed, versioned, and citable.
the canonical record
Built for four readers, in balance.
No one reader displaces another. The human reading experience is first-class; machine-readability is additive, never a substitute.
Authors
Publish work that confers status — affordably, and without friction. No fee, no gatekeeping tax.
Reviewers
Review in a structured tool that respects your time. Every review is signed, visible, and credited.
Human readers
A clean, canonical, beautifully rendered record of science — precise to cite, easy to verify.
Agentic readers
Every paper as a typed API — claims, evidence, and citations as objects, not prose.
GET /v1/papers · application/opus.v1+jsonA journal with the ambition of Nature and the engineering of Stripe.
One record, every reader. Add your work to the canonical record of science — or review what others have submitted.