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.
Gemini 2.0 Pro is Google's most advanced model to date, delivering exceptional coding performance and handling complex prompts with ease. It comes equipped with enhanced capabilities such as native tool integration, image generation, and speech synthesis. Designed for advanced reasoning, the model supports multimodal inputs, including text, images, video, and audio. Available via Google AI Studio and Vertex AI, it offers substantial performance improvements over previous versions while maintaining high efficiency.
| Qwen2.5-VL-32B | Gemini 2.0 Pro | |
|---|---|---|
Provider | ||
Web Site | ||
Release Date | Mar 25, 2025 4 weeks ago | Dec 11, 2024 4 months ago |
Modalities | text images video | text images voice video |
API Providers | - | Google AI Studio, Vertex AI |
Knowledge Cut-off Date | Unknown | 08.2024 |
Open Source | Yes (Source) | No |
Pricing Input | $0 | $0.10 per million tokens |
Pricing Output | $0 | $0.40 per million tokens |
MMLU | 78.4% Source | Not available |
MMLU Pro | 49.5% | 79.1% Source |
MMMU | 70% | 72.7% Source |
HellaSwag | Not available | Not available |
HumanEval | Not available | Not available |
MATH | 82.2% | 91.8% Source |
GPQA | 46.0% Diamond | 64.7% Diamond Source |
IFEval | Not available | Not available |
Mobile Application | - |
Compare AI. Test. Benchmarks. Mobile Apps Chatbots, Sketch
Copyright © 2025 All Right Reserved.