Amazon Nova 2 Lite Leads Video Understanding Benchmark With Unmatched Price Performance

Amazon Web Services

Apr 8, 2026 · Updated Jun 5, 2026

Amazon Nova 2 Lite achieved the top spot for price-performance in the MediaPerf video understanding benchmark. While ranking second in overall quality, the model delivered the lowest operational cost at just $0.54 per hour of video. This shift makes large-scale video tagging economically viable for media companies.

Amazon's Nova 2 Lite model emerged as the leader for video understanding in the latest MediaPerf benchmark from Coactive AI, a multimodal (AI that processes multiple data types together) platform. The model achieved an F1 macro score (a metric balancing precision and recall) of 0.377, ranking second in quality while maintaining the lowest cost tested.

Video tagging is traditionally one of the most expensive AI workloads. Nova 2 Lite disrupts this by outperforming significantly larger models, including Gemini 3.0 Pro and Amazon's own Nova Pro, in both accuracy and efficiency. It allows companies to process massive video archives without the high costs of frontier models.

You can now deploy Nova 2 Lite for production video tagging and content moderation via Amazon Bedrock. At $0.54 per hour of video, it is the default choice for high-volume pipelines requiring near-frontier accuracy at roughly 25% of the cost of GPT 5.1.

View the full update on mediaperf.org

AWS AI

@AWSAIApr 7

Amazon Nova 2 Lite earns the top spot in video understanding. MediaPerf from @CoactiveAI evaluates multimodal models for video understanding tasks that media and entertainment customers run in production. https://t.co/zPgK6Qxsmc

View on X

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

See all AI news & updates from AWS →

Keep reading

Arena Ranks Google Gemma 4 as Top Open Vision Model

Google's Gemma-4-31b and Gemma-4-26b-a4b have entered the Vision Arena leaderboard as the #2 and #4 ranked open models. These releases shift the price-performance frontier by delivering vision reasoning capabilities that rival proprietary systems at a fraction of the cost.

GoogleApr 1

Google releases Veo 3.1 Lite for high volume video at half cost

Google launched Veo 3.1 Lite, a new video generation model available through the Gemini API and Google AI Studio that supports text-to-video and image-to-video workflows. By offering the same speed as the Fast model at less than half the price, this update makes high-volume video applications economically viable for developers.

OpenRouter Hosts Perceptron Mk1 for Structured Video and Embodied Reasoning

OpenRouterMay 12

OpenRouter Hosts Perceptron Mk1 for Structured Video and Embodied Reasoning

OpenRouter integrated Perceptron Mk1, a vision-language model designed for spatial grounding and video understanding with structured outputs like bounding boxes and timestamps. The model introduces a reasoning toggle for complex visual tasks at a significantly lower price point than general-purpose frontier models.

LumaMay 5

Luma Launches Uni-1.1 API With Built-In Prompt Intelligence and Lower Costs

Luma released the Uni-1.1 API, a video and image generation endpoint featuring integrated prompt enhancement and reference gathering. The model targets production-grade workflows by offering significantly lower latency and pricing than existing frontier alternatives.