Nova Lite: benchmark results
Amazon's low-cost multimodal Bedrock model (December 2024) taking text, images, and video in a 300K context. Provider: Amazon. Released 2024-12-03. Access: API.
Unified ELO 1480 ± 1, rank #786 of 1392 rated models, from 86 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HELM NaturalQuestions (Open) | 81.52 | F1 (%) | 98.9 |
| Ko-AgentBench - L6 Efficient Tool Utilization | 31.33 | Efficiency Score (%) | 92.9 |
| HELM NarrativeQA | 76.81 | F1 (%) | 81.1 |
| HELM (Stanford) | 70.8 | Mean Win Rate (%) | 77.8 |
| Translation en to Set1 COMET22 | 88.8 | en→Set1 COMET22 (self-reported) | 66.7 |
| Translation en to Set1 spBleu | 41.5 | en→Set1 spBLEU (self-reported) | 66.7 |
| HELM WMT 2014 | 20.43 | BLEU-4 (%) | 65 |
| LLM Stats (DROP) | 80.2 | Score (%) | 64.3 |
| LLM Stats (ChartQA) | 86.8 | Score (%) | 64 |
| Translation Set1 to en COMET22 | 88.8 | Set1→en COMET22 (self-reported) | 62.5 |
| Translation Set1 to en spBleu | 43.1 | Set1→en spBLEU (self-reported) | 58.3 |
| AA Omniscience - Software Engineering (SWE) - Julia | 8.33 | Accuracy (%) | 52.6 |
Interactive version: theaggregate.ai/model?slug=nova-lite · How It Works · Data refreshed daily, snapshot 2026-09-05.