Week 26ARCHIVED · JUN 22–28

SciWorkflow

Long-Horizon Agentic Benchmark

SciWorkflow@paulp
  • AI
  • Analytics
  • Web apps
On the web

About this project

Pitch

SciWorkflow is built for science rather than software. Instead of simple one-step answers or code patches, we evaluate complex 20 to 30 step long horizon workflows specifically in biology and materials science, addressing a massive unmet need for hard scientific validation.

Why we're launching this

From the team
Current industry benchmarks like SWE bench focus entirely on software engineering code fixes. General assistant benchmarks like GAIA focus on basic everyday web tasks. Differentiation: SciWorkflow is built for science rather than software. Instead of simple one-step answers or code patches, we evaluate complex 20 to 30 step long horizon workflows specifically in biology and materials science, addressing a massive unmet need for hard scientific validation.
SciWorkflow

@paulp · founder

Discussion

0 comments

Sign in to join the discussion.

No takes yet. Be the first to weigh in.

More in AI

See all →