0

Attribution

Where Sophon's data comes from.

Sophon is a derivative work - every artefact in the catalog originates with someone else. We attribute on every page where we render hosted content (look for the source badge under READMEs, abstracts, and model cards). This page lists the upstream data sources we ingest from.

Papers, code & metadata

arXiv ↗Paper metadata, abstracts, license detection
Semantic Scholar ↗TLDR summaries, citation graph, author affiliations
Papers With Code ↗Paper ↔ code ↔ benchmark links (HuggingFace-run revival)
GitHub ↗README files, license detection, canonical repos
Hugging Face ↗Model cards, dataset samples, model metadata

Eval & RL-environment registries

Inspect AI ↗Curated eval definitions with author + paper provenance
Prime Intellect ↗RL environment registry
Atropos (Nous Research) ↗RL environment registry
OpenEnv (Meta PyTorch) ↗RL environment registry
OpenReward (General Reasoning) ↗RL environment & reward aggregator
Harbor (Laude Institute) ↗Agent-eval framework & RL-environment registry

Leaderboards & benchmark scores

LMArena ↗Human-preference Elo ratings
Artificial Analysis ↗Aggregated intelligence-index scores
OpenRouter ↗Model metadata, pricing, and availability
Open LLM Leaderboard ↗Open-model benchmark scores
SWE-bench ↗Coding-agent benchmark scores
LiveBench ↗Contamination-resistant benchmark scores
Aider ↗Code-editing (polyglot) leaderboard
EvalPlus ↗HumanEval+ / MBPP+ code scores
Humanity's Last Exam ↗Frontier-knowledge benchmark scores
MathArena ↗Competition-math benchmark scores
FrontierMath (Epoch AI) ↗Research-math benchmark scores
AlpacaEval ↗Instruction-following leaderboard
Arena-Hard ↗Hard-prompt, LLM-judge scores
OSWorld ↗Computer-use agent benchmark scores
τ-bench ↗Tool-agent benchmark scores
TruthfulQA ↗Truthfulness benchmark scores
DABstep ↗Data-analysis agent benchmark scores
AfterQuery ↗IDE-Bench, Market Bench, App-Bench & FinanceQA scores
Vals AI ↗Proprietary domain benchmarks (tax, finance, legal, medical)