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 Flash is Google's high-performance, low-latency model designed to drive advanced agentic experiences. Equipped with native tool integration, it supports multimodal inputs, including text, images, video, and audio. Offering substantial improvements over previous versions, the model balances efficiency, speed, and enhanced capabilities for seamless real-time interactions.
| Qwen2.5-VL-32B | Gemini 2.0 Flash | |
|---|---|---|
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% | 77.6% Source |
MMMU | 70% | 71.7% Source |
HellaSwag | Not available | Not available |
HumanEval | Not available | Not available |
MATH | 82.2% | 90.9% Source |
GPQA | 46.0% Diamond | 60.1% Diamond Source |
IFEval | Not available | Not available |
Mobile Application | - |
Compare AI. Test. Benchmarks. Mobile Apps Chatbots, Sketch
Copyright © 2025 All Right Reserved.