GIM: Grounded Integrated Multitask Benchmark for AI Evaluation

GIM is a benchmark dataset for evaluating AI models, created by Facebook. It focuses on tasks that integrate multiple cognitive domains to avoid conflating memorization with capability or divorcing reasoning from practical contexts. The dataset was last updated on May 19, 2026.

Use Cases

• Benchmarking AI models on tasks that require integrated cognitive skills based on the dataset's stated purpose
• Evaluating reasoning in practical, grounded contexts as described in the dataset's approach
• Comparing model performance against benchmarks that escalate knowledge demands or focus on abstract reasoning
• Developing new evaluation metrics for AI based on multi-domain cognitive tasks

Strengths

• Created by Facebook, a major AI research organization
• Designed to address specific limitations of existing LLM benchmarks as described in the summary

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download
• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

facebook

Freshness

Last updated 2026-05-19 14:55:23