Open model benchmark leaderboards

Task-specific results for open-weight models and reproducible open-source systems. Every score links to the original paper, official leaderboard, or first-party result file.

Model scores and agent-system scores are labeled separately. Results from different benchmark versions, splits, tools, or step budgets should not be treated as directly comparable.
Abstract reasoning

ARC-AGI

Abstract reasoning benchmark built from novel visual transformation tasks. Results depend heavily on the dataset split and solver system, so each row preserves its evaluation scope.

Metric
Solved tasks
Rows
3 sourced results
 Retrieval-augmented generation

CRAG

Comprehensive RAG benchmark for factual question answering across dynamic facts, long-tail entities, web pages, and mock knowledge-graph APIs.

Metric
Answer accuracy
Rows
4 sourced results
 Agent memory

LongMemEval

Long-term conversational memory benchmark covering information extraction, multi-session reasoning, knowledge updates, temporal reasoning, and abstention.

Metric
End-to-end QA accuracy at top-10 retrieval
Rows
2 sourced results
 Computer use

OSWorld-Verified

Real-computer benchmark for multimodal agents completing open-ended tasks across desktop and web applications.

Metric
Task success rate
Rows
4 sourced results

How scores enter this database

We preserve the benchmark version, evaluation scope, system setup, openness, and verification status for every row. A result is omitted when its exact score or provenance cannot be confirmed from a primary source.