Nova 2.0 Lite (High): benchmark results

Provider: Amazon. Released 2025-12-02. Access: API.

Unified ELO 1563 ± 1, rank #601 of 2033 rated models, from 104 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA AIME 202594.33Accuracy (%)95.4
Nejumi 4 - Toxicity - Prohibited Acts97.92Criteria met (%)93.4
Nejumi 4 - MT-Bench (Japanese) - Humanities100Judge rating (1-10, x10)90.4
Nejumi 4 - MT-Bench (Japanese) - STEM100Judge rating (1-10, x10)90.4
Nejumi 4 - MT-Bench (Japanese) - Math100Judge rating (1-10, x10)89
AA IFBench70.75Accuracy (%)85.6
AA LiveCodeBench71.11Pass@1 (%)81.4
AA MMLU-Pro81.82Accuracy (%)76.7
Nejumi 4 - JBBQ (2-shot) - Accuracy92Accuracy (%)72.8
Nejumi 4 - MT-Bench (Japanese) - Roleplay98Judge rating (1-10, x10)72.4
AA GPQA Diamond81.11Accuracy (%)69.1
Nejumi 4 - jaster (0-shot) - MAWPS98Exact match (%)68.4

Interactive version: theaggregate.ai/model?slug=nova-2-0-lite-high · How It Works · Data refreshed daily, snapshot 2026-09-28.