Over the past five months since the release of Qwen2-VL, developers have built new models based on it, contributing valuable feedback. Now, Qwen2.5-VL introduces enhanced capabilities, including precise analysis of images, text, and charts, as well as object localization with structured JSON outputs. It understands long videos, identifies key events, and functions as an agent, interacting with tools on computers and phones. The model's architecture features dynamic video processing and an optimized ViT encoder for improved speed and accuracy.
Amazon Nova Pro is a cutting-edge multimodal model designed to handle text, image, and video inputs with advanced processing capabilities. Featuring a 300K-token context window, it excels in document analysis, visual question answering, and complex agent-driven workflows. As part of the Amazon Nova foundation models, it supports fine-tuning and distillation, enabling deep customization for various applications.
| Qwen2.5-VL-32B | Nova Pro | |
|---|---|---|
Provider | ||
Web Site | ||
Release Date | Mar 25, 2025 4 weeks ago | Dec 02, 2024 4 months ago |
Modalities | text images video | text images video |
API Providers | - | Amazon Bedrock |
Knowledge Cut-off Date | Unknown | Purposefully not disclosed |
Open Source | Yes (Source) | No |
Pricing Input | $0 | $0.80 per million tokens |
Pricing Output | $0 | $3.20 per million tokens |
MMLU | 78.4% Source | 85.9% CoT Source |
MMLU Pro | 49.5% | Not available |
MMMU | 70% | Not available |
HellaSwag | Not available | Not available |
HumanEval | Not available | 89% pass@1 Source |
MATH | 82.2% | 76.6% CoT Source |
GPQA | 46.0% Diamond | 46.9% Main Source |
IFEval | Not available | 92.1% Source |
Mobile Application | - | - |
Compare AI. Test. Benchmarks. Mobile Apps Chatbots, Sketch
Copyright © 2025 All Right Reserved.