Nova 2 Lite
Amazon Nova 2 Lite is a low-latency, cost-effective multimodal reasoning model for everyday workloads. It processes text, images, and video to generate text output, with hybrid reasoning controls that let developers toggle extended thinking on or off and adjust depth (low, medium, high) to balance accuracy, speed, and cost. It supports up to 1M tokens of context and matches or exceeds Claude Haiku 4.5, GPT-5 Mini, and Gemini 2.5 Flash on most standard benchmarks.
Benchmark results
| Benchmark | Score | Tags | Source |
|---|---|---|---|
| AIME 2025 | 91.0% | self-reported llm-stats | link → |
| BFCL-V4 | 60.3% | self-reported llm-stats | link → |
| GPQA | 79.6% | self-reported llm-stats | link → |
| IFBench | 70.8% | self-reported llm-stats | link → |
| LiveCodeBench | 71.0% | self-reported llm-stats | link → |
| LongCodeBench | 84.0% | self-reported llm-stats | link → |
| MCP Atlas | 24.6% | self-reported llm-stats | link → |
| MMLU-Pro | 80.9% | self-reported llm-stats | link → |
| MMMU-Pro | 61.8% | self-reported llm-stats | link → |
| Multi-Challenge | 76.6% | self-reported llm-stats | link → |
| OCRBench_V2 | 56.1% | self-reported llm-stats | link → |
| QVHighlights | 77.2% | self-reported llm-stats | link → |
| RealKIE-FCC | 62.1% | self-reported llm-stats | link → |
| ScreenSpot | 83.3% | self-reported llm-stats | link → |
| SWE-Bench Verified | 64.5% | self-reported llm-stats | link → |
| Tau2 Airline | 64.8% | self-reported llm-stats | link → |
| Tau2 Retail | 76.5% | self-reported llm-stats | link → |
| Tau2 Telecom | 76.0% | self-reported llm-stats | link → |
| Terminal-Bench | 32.5% | self-reported llm-stats | link → |