LetTokenBurn
Sign in
Verified86%Harbor: pendingAgentic / Tool UseKnowledge & Reasoning

PhysClosureBench

From Research Goals to Scientific Closure: Evaluating Agents Through Evidence-Supported Physics Research

PhysClosureBench — Complete 50-Case Materials Paper This collection provides the task specifications and complete case-specific grading materials for all 50 PhysClosureBench research tasks.

PaperRepoDataset

How to run

ltb · physclosurebench
$ npm i -g lettokenburn-cli
$ ltb bench info physclosurebench

External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.

Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs

Scores

7 sourced entries
HarnessSourceConfidenceLink
Claude Opus 56.3N/AIndependentmediumOct 15, 2026arxiv.org
GPT-5.6 Sol5.8N/AIndependentmediumOct 15, 2026arxiv.org
Kimi K35.2N/AIndependentmediumOct 15, 2026arxiv.org
DeepSeek V4.1 Flash4.5N/AIndependentmediumOct 15, 2026arxiv.org
Qwen3.8 Max3.5N/AIndependentmediumOct 15, 2026arxiv.org
GLM-5.3 Flash3.3N/AIndependentmediumOct 15, 2026arxiv.org
Gemini 3.1 Pro2.9N/AIndependentmediumOct 15, 2026arxiv.org

Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.

Reproduced by the community

All runs
Loading community runs…

About

PhysClosureBench — Complete 50-Case Materials

Paper This collection provides the task specifications and complete case-specific grading materials for all 50 PhysClosureBench research tasks. Each case directory contains task.md, grading_rubric.md, grading_rubric.json, resource_limits.json, and sources.md. The Markdown and JSON rubrics contain the same twenty half-point criteria, task-specific RC0–RC4 requirements, scientific dependencies, accepted equivalent evidence, and…