# GaRAGe

> Amazon benchmark of grounded answers in retrieval-augmented generation, with human labels for which supplied passages are relevant.

Canonical page: https://benchlm.ai/benchmarks/garage

- Category: [Knowledge](/knowledge)
- Last updated: October 29, 2026

## About GaRAGe

- Year: 2025
- Tasks: Questions with retrieved passages
- Format: Relevance-aware factuality
- Difficulty: Facts from supplied sources
- Paper: [GaRAGe](https://github.com/amazon-science/GaRAGe)

NVIDIA reports it for Nemotron 3.5 Bolt E4B (October 23, 2026), run in its NeMo Gym harness, as the relevance-aware factuality score in reasoning mode.

GaRAGe is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (2 models)

| Rank | Model | Creator | Score |
|------|-------|---------|-------|
| 1 | [Nemotron 3.5 Bolt E4B BF16](/models/nemotron-3-5-bolt-e4b-bf16) | NVIDIA | 42.61% |
| 2 | [Nemotron 3.5 Bolt E4B](/models/nemotron-3-5-bolt-e4b) | NVIDIA | 40.73% |

## FAQ

### What does GaRAGe measure?

Amazon benchmark of grounded answers in retrieval-augmented generation, with human labels for which supplied passages are relevant.

### Which model scores highest on GaRAGe?

Nemotron 3.5 Bolt E4B BF16 by NVIDIA currently leads with a score of 42.61% on GaRAGe.

### How many models are evaluated on GaRAGe?

2 AI models have published results on GaRAGe in the BenchLM catalog.

### Does GaRAGe affect BenchLM's overall score?

Not directly. GaRAGe is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Compare Top Models on GaRAGe

- [Nemotron 3.5 Bolt E4B BF16 vs Nemotron 3.5 Bolt E4B](/compare/nemotron-3-5-bolt-e4b-vs-nemotron-3-5-bolt-e4b-bf16)