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.
The OpenAI o3-mini is a high-speed, cost-effective reasoning model designed for STEM applications, with strong performance in science, mathematics, and coding. Launched in January 2025, it includes essential developer features such as function calling, structured outputs, and developer messages. The model offers three reasoning effort levels—low, medium, and high—allowing users to optimize between deeper analysis and faster response times. Unlike the o3 model, it lacks vision capabilities. Initially available to select developers in API usage tiers 3-5, it can be accessed via the Chat Completions API, Assistants API, and Batch API.
Qwen2.5-VL-32B | o3-mini | |
---|---|---|
Provider | ||
Web Site | ||
Release Date | Mar 25, 2025 4 weeks ago | Jan 31, 2025 2 months ago |
Modalities | text images video | text |
API Providers | - | OpenAI API |
Knowledge Cut-off Date | Unknown | Unknown |
Open Source | Yes (Source) | No |
Pricing Input | $0 | $1.10 per million tokens |
Pricing Output | $0 | $4.40 per million tokens |
MMLU | 78.4% Source | 86.9% pass@1, high effort Source |
MMLU Pro | 49.5% | Not available |
MMMU | 70% | Not available |
HellaSwag | Not available | Not available |
HumanEval | Not available | Not available |
MATH | 82.2% | 97.9% pass@1, high effort Source |
GPQA | 46.0% Diamond | 79.7% 0-shot, high effort Source |
IFEval | Not available | Not available |
Mobile Application | - |
Compare AI. Test. Benchmarks. Mobile Apps Chatbots, Sketch
Copyright © 2025 All Right Reserved.