Back to Ecosystem
ResearchClawBench logo
Research & BenchmarksVerified

ResearchClawBench

An end-to-end benchmark for autonomous scientific research agents, covering real datasets, code, figures, reports, and peer-review-style evaluation across multiple disciplines.

Decision information

Verified: 8/14/2026
Ecosystem category
Research & Benchmarks
Provider
InternScience
Content type
Research
Pricing
Open Source
Open source
Yes
License
MIT
Maintenance
Active
Install methods
Source
Platforms
Linux
Interfaces
Web, Terminal

Security signals (informational, not a security certification)

Source availableRequires shellRequires credentials

Verification source: https://arxiv.org/abs/2606.07591 · arXiv:2606.07591

Overview

An end-to-end benchmark for autonomous scientific research agents, covering real datasets, code, figures, reports, and peer-review-style evaluation across multiple disciplines.

What it provides

  • Forty real-science tasks spanning ten disciplines
  • Agent-agnostic execution with multimodal rubric scoring
  • Supports OpenClaw, ResearchClaw, coding agents, and custom harnesses

AllClaw classification

This project supports the Claw ecosystem as research benchmarks; it is not presented as a standalone Claw agent.

Provider: InternScience. Primary source: https://arxiv.org/abs/2606.07591.

The classification and core claims were checked against the linked primary source on August 14, 2026.

Tags

benchmarkscientific-researchevaluationmulti-agentleaderboard
ResearchClawBench | AllClaw | All Claw