Claude Science
Local code, scientific connectors, HPC/GPU orchestration, and artifact provenance.
Research platform benchmark
Compare the workflow,not the marketing.
A dated, source-backed view of what scientific research platforms document today: evidence access, full text, computation, private data, formal review, and provenance.
11Platforms reviewed
6Capabilities scored
NoPaid placement or hidden weighting
Coverage across six audited capabilities
Full color means documented. Hatched means partial or gated. This is not an answer-quality score.
DocumentedPartial / gated
The number at right is documented + partial/gated coverage. It is a compact reading aid, not a combined rank: six different capabilities are not equally valuable to every buyer.
Category leaders
Different products leaddifferent kinds of work.
Cortexa does not appear below. We do not award ourselves a category on our own comparison page, and no common-task benchmark exists across this market that would settle one.
Elicit
Protocol, screening, extraction, dual review, and PRISMA reconstruction.
Edison / Kosmos
Long autonomous runs over literature and analysis code.
Asta
Open agents, scholarly graph, notebooks, and a public benchmark.
Scite
Full-text corpus and citation-statement classification.
Consensus
Publishes exact deep-review counts and prices a buyer can do arithmetic on.
OpenAI
Specialist life-science models and plugins across genomics, chemistry, and lab work.
Co-Scientist and ERA, with peer-reviewed validation in Nature.
Where Cortexa is different
One hosted path from evidence to delivery.
Cortexa is the broadest hosted, low-friction research surface: public evidence across many disciplines, private files and connections, bounded managed computation, a reusable file library, continuity across web, API, MCP and Slack, and editable shareable artifacts — with transparent credit accounting. That combination is rare.
It is not the leader in formal systematic review, licensed full-text coverage, local and HPC execution provenance, autonomous discovery depth, or enterprise certification.
See the science workflowsPublic evidence
Literature, trials, regulatory sources, patents, and live web research.
Private context
Selected files, databases, storage, APIs, Slack, and GitHub.
Managed computation
Python, long-running analysis, and cost-bounded background work.
Editable delivery
Research abstracts, presentations, dashboards, spreadsheets, and shared views.
Full matrix
The evidence underneath the chart.
| Platform | Sources | Full text | Code | Private data | Formal review | Provenance |
|---|---|---|---|---|---|---|
| Cortexa Hosted multi-discipline research, private files and connections, managed computation, and editable shareable artifacts. | Documented | Not documented | Documented | Documented | Not documented | Partial |
| Claude Science Local scientific workbench with code, connectors, HPC and GPU orchestration, and deep artifact provenance. | Documented | Partial | Documented | Documented | Not documented | Documented |
| Elicit Structured literature workflows and the strongest documented formal systematic-review process in this set. | Documented | Partial | Partial | Documented | Documented | Documented |
| Edison / Kosmos Long autonomous discovery runs combining large literature reads and extensive analysis code. | Documented | Unknown | Documented | Gated | Not documented | Documented |
| Asta Open, reproducible scientific agents, a large scholarly graph, and notebook-oriented workflows. | Documented | Documented | Documented | Partial | Not documented | Partial |
| SciSpace Broad paper discovery, reading, extraction, writing, and literature-review assistance. | Documented | Partial | Partial | Partial | Partial | Partial |
| Consensus Literature-grounded narrative answers with unusually transparent retail usage and pricing. | Documented | Partial | Not documented | Partial | Not documented | Partial |
| Scite Full-text literature and citation-context infrastructure at exceptional scale. | Documented | Documented | Not documented | Partial | Not documented | Partial |
| OpenAI science stack General research and code execution plus specialist life-science model and plugin work. | Documented | Unknown | Documented | Documented | Not documented | Partial |
| Google science stack Deep Research, Co-Scientist, and specialized scientific systems with strong task-level validation. | Documented | Unknown | Documented | Documented | Not documented | Partial |
| Perplexity General deep research, enterprise search, files, and code-oriented analysis. | Partial | Unknown | Documented | Documented | Not documented | Partial |
Methodology
What the labels mean
Documented means the vendor clearly describes the capability. Partial means narrower support or an incomplete workflow. Not documented and unknown are not claims that a feature is impossible; they mean we found no current public contract.
What this does not measure
These labels do not prove implementations are equally good, and they do not compare answer quality. There is no independent common-task product benchmark across this market today.
Vendor documentation and official announcements checked 17 August 2026. Cortexa capabilities rechecked against the running product and repository 25 August 2026. Product names belong to their respective owners. Corrections: support@cortexa.sh.
The honest benchmark is the one you can inspect.
We will update this page as products change and replace capability labels with common-task measurements when those measurements exist.