Research AI Landscape

kosmos

2 min read

Kosmos

Status: watch-list | Maturity: experimental | Last reviewed: 2026-03-04

What it does

Kosmos is an “AI Scientist” developed by FutureHouse (now managed by Edison Scientific) that performs extensive literature review and analysis to generate novel research hypotheses. A single Kosmos run reads approximately 1,500 papers and runs 42,000 lines of analysis code using structured world models to synthesise information across hundreds of agent trajectories.

Key capabilities

  • Large-scale automated literature review (approximately 1,500 papers per run)
  • Novel hypothesis generation grounded in synthesised literature
  • Structured world models for coherent synthesis across tens of millions of tokens
  • Targeted use for specific, well-defined research objectives

Fit for Sheffield researchers

Currently experimental and expensive ($200 per run). Best suited to targeted use by senior researchers with specific, well-defined research questions where hypothesis generation from large literature corpora adds clear value. Not appropriate for routine or exploratory research at current pricing.

Licensing and access

Pay-per-run model at $200 per run. No institutional pricing at time of review. Recommend monitoring as the product matures before considering institutional access.

Trust and governance notes

Insufficient information on data handling, GDPR compliance, or data residency. High cost per run also creates governance questions about research budget approval processes.

Observed limitations

  • $200 per run is prohibitive for routine research use
  • Experimental maturity — output quality not independently validated at scale
  • Corpus limited primarily to STEM literature
  • Novel hypotheses require extensive expert scientific validation
  • Not suitable for arts, humanities, or qualitative research

paperqa2 · crow · google-co-scientist

Sources and evidence

Graph View

Backlinks