Sign in Subscribe
2 min read

OpenAI’s GPT-Image-1.5 Dethrones Nano Banana Pro as Meta’s SAM Audio Redefines Sound Editing

New AI Models & Releases

OpenAI Releases GPT-Image-1.5 (ChatGPT Images), Dominating LMArena Benchmarks
OpenAI’s new GPT-Image-1.5 model has claimed the #1 spot on LMArena, outperforming Google’s Nano Banana Pro. The model features major improvements in speed, editing capabilities, and consistency, and is now accessible via a dedicated "Images" tab in ChatGPT and as an API.

Microsoft Launches TRELLIS 2-4B: Open-Source Image-to-3D Model
Microsoft’s TRELLIS 2-4B is an open-source model that converts single images into 3D assets using Flow-Matching Transformers and Sparse Voxel-based 3D VAE. The model is available on Hugging Face with a demo and technical blog.

QwenLong-L1.5 Achieves State-of-the-Art Long-Context Reasoning (4M Tokens)
QwenLong-L1.5 introduces breakthroughs in long-context AI with novel data synthesis, stabilized RL, and memory management, supporting contexts up to 4M tokens. The model is available on Hugging Face.

Meta Unveils SAM Audio Model for Advanced Sound Segmentation
Meta’s new SAM Audio Model enables precise sound isolation from complex audio mixtures using text, visual, and time-span prompts, revolutionizing audio editing workflows.

Fine-Tuned Qwen3-VL-30B-A3B for Browser Automation (bu-30b-a3b-preview)
A specialized fine-tuned version of Qwen3-VL-30B-A3B, named browser-use/bu-30b-a3b-preview, excels in browser-based tasks, delivering high efficiency and reliability at low compute costs (~$1 for hundreds of tasks).

Mistral AI Introduces "Mistral Small Creative" for Creative Writing & Roleplay
Mistral’s new experimental model, Mistral Small Creative, is optimized for creative writing, narrative generation, and conversational agents, supporting a 32k context length. It is available via the Mistral API under the Labs program.

Platform & Service Updates

OpenAI Rolls Back ChatGPT’s Model Router System for Most Users
OpenAI has reverted its model router system, which previously directed some queries to a "Thinking" model. Free users will now default to GPT-5.2 Instant but can manually opt into the Thinking model.