• 7 min read
Vision Token Counts: What does it cost to process an image with a frontier vision model?
Understand the cost, per-provider tokenization rules, and a comparison across image sizes for Claude, GPT, and Gemini.
• 5 min read
Which is the Best Coding Agent for Vision tasks?
SUMMARY
This benchmark pits four coding agents (Claude Code with Opus 4.6, Cursor with Composer 2, Gemini CLI with Gemini 3.1 Pro, and Codex with GPT 5.4) against five computer vision tasks including bird counting with SAM 3, car counting in video and RTSP streams, avocado detection,
• 7 min read
Best Multimodal Models in 2026
From SAM 3’s record-breaking segmentation speed to Gemini 3’s massive 2-million-token context window, explore the top models that can "see," reason, and deploy in production today.
• 6 min read
Launch: Use Segment Anything 3 (SAM 3) with Roboflow
SUMMARY
Roboflow has integrated SAM 3, Meta's foundation model for Promptable Concept Segmentation, across every major part of its platform, covering the Playground for interactive testing, the inference API, edge and private-cloud deployment, Roboflow Workflows for building production pipelines, and the Label Assist annotation tool for accelerating
• 9 min read
How to Detect, Track, and Identify Basketball Players with Computer Vision
Build a computer vision pipeline that detects, tracks, and identifies NBA players during a game. Using models like RF-DETR, SAM2, SigLIP, SmolVLM2, and ResNet, the system handles motion blur, occlusion, and uniform similarity to reliably map jersey numbers to player identities.
• 10 min read
How to Fine-Tune Qwen2.5-VL with a Custom Dataset
Learn how to fine-tune Qwen2.5-VL for document processing using a custom dataset.
• 5 min read
GPT-5 for Vision: Results from 80+ Real-World Tests
SUMMARY
Roboflow tested GPT-5 across more than 80 structured visual tasks spanning document understanding, defect detection, object counting, and the RF100-VL object detection benchmark. On Vision Checkup, GPT-5 tied for first place with GPT o4 Mini, driven largely by strong spatial reasoning, but scored only 4 of
• 5 min read
Use Qwen2.5-VL for Zero-Shot Object Detection
SUMMARY
Qwen2.5-VL is an open source vision-language model available in 3B, 7B, and 72B sizes that can locate and label objects in an image from a plain-text prompt, with no labeled training data or fine-tuning required. This tutorial runs the 7B variant on a T4
• 7 min read
How to Fine-Tune a SmolVLM2 Model on a Custom Dataset
Learn how to fine-tune SmolVLM2 for visual question answering using a custom dataset.
• 4 min read
OpenAI o3-pro: Multimodal and Vision Analysis
Explore how OpenAI o3-pro does on a range of use cases, from defect detection to object counting to VQA.
• 6 min read
OpenAI o3 and o4-mini: Multimodal and Vision Analysis
Read our analysis of how OpenAI's O3 and O4-Mini models perform on a range of vision tasks.
• 6 min read
OpenAI GPT-4.1: Multimodal and Vision Analysis
Read our analysis of OpenAI's GPT-4.1 model on multimodal tasks like VQA, object detection, and more.
• 6 min read
Gemma 3: Multimodal and Vision Analysis
Read our analysis of Google's Gemma-3 model and how it performs on a variety of common multimodal vision AI tasks.
![Foundational Few-Shot Object Detection Challenge [CVPR 2025]](https://storage.ghost.io/c/2c/8d/2c8d8c0d-1c15-4b6d-825e-02b78d61d40a/content/images/size/w640/2025/03/img-foundational-few-shot-object-detection-challenge-1.png)
• 2 min read
Foundational Few-Shot Object Detection Challenge [CVPR 2025]
Roboflow & Carnegie Mellon University are releasing the second iteration of the Foundational Few-Shot Object Detection Challenge at CVPR 2025.
• 5 min read
SmolVLM2: Multimodal and Vision Analysis
Read our analysis of how SmolVLM2 performs on a range of multimodal vision tasks.
• 6 min read
Moondream 2: Multimodal and Vision Analysis
Read our analysis of how the multimodal Moondream 2 model performs on a range of vision tasks.
• 8 min read
OpenAI o3-mini: Vision and Multimodal Features
Read our analysis of how OpenAI's O3 Mini model performs on various computer vision tasks.
• 8 min read
What is Few-Shot Learning?
In this blog post, we discuss what few-shot learning is, architectural approaches for implementing few-shot learning, and specific implementations of few-shot learning techniques.
• 13 min read
How to Fine-tune PaliGemma 2
Learn how to fine-tune PaliGemma 2 to extract data from an image in JSON format.
• 7 min read
Launch: Use Florence-2 in Roboflow Workflows
Learn how to use Florence-2 in Roboflow Workflows for zero-shot object detection, OCR, and more.
• 7 min read
Table and Figure Understanding with Computer Vision
Learn how to use Roboflow Workflows and multimodal models to derive information about the contents of tables and figures.
• 15 min read
What is CLIP? Contrastive Language-Image Pre-Training Explained.
CLIP is an open source, multimodal computer vision model developed by OpenAI. Learn what CLIP is in this guide.
• 7 min read
How to OCR Hand-Written Notes with GPT-4
Learn how to OCR hand-written notes with GPT-4.
• 4 min read
Document Understanding with Multimodal Models
Learn how to use the PaliGemma multimodal model to ask questions about the contents of a document.
Stay Connected