Alibaba’s Qwen3-Omni & Lucy-Edit Redefine Multimodal AI as Grok 4 Fast Outperforms Gemini at 25x Lower Cost

New AI Models

Qwen3-Omni (7B) Released: Alibaba’s new multimodal AI model, Qwen3-Omni, supports text, images, video, and audio with advanced multilingual and reasoning capabilities. It is designed for high-performance applications across diverse tasks.

• Qwen3Omni
• Qwen3-Omni, Qwen/Qwen3-Omni-7B spotted
  • Hugging Face Transformers PR

Qwen3-4B Fine-Tuned for Function Calling: A lightweight version of Qwen3-4B optimized for coding assistance, requiring only 6GB of VRAM. It supports tool-calling and integrates with frameworks like Ollama and Codex.

• Just dropped: Qwen3-4B Function calling on just 6GB VRAM
  • Hugging Face Model
  • Open-Codex GitHub
  • Anything-Codex GitHub
  • Anon-Codex GitHub

Lucy-Edit: First Open-Source Video Editing Model: Based on Wan2.2 5B, Lucy-Edit-Dev enables AI-powered video modifications, including changes to clothes, characters, backgrounds, and objects.

• Lucy-Edit: 1st Open-sourced model for Video editing
  • Hugging Face Model

Benchmarks & Performance

Grok 4 Fast Matches Gemini 2.5 Pro at 25x Lower Cost: ArtificialAnalysis benchmarks suggest Grok 4 Fast delivers comparable intelligence to Google’s Gemini 2.5 Pro while being significantly more cost-efficient.

• ArtificialAnalysis claims Grok 4 Fast matches Gemini 2.5 Pro's intelligence at 25x lower cost
  • Tweet by Koltregaskes

Developer Tools & Frameworks

Vogte: Open-Source Agentic TUI for Go: A new terminal-based interface tool for Go developers, enabling agentic workflows and customization for command-line tasks.

• Vogte: Open Source Agentic TUI for Go