model-release

OpenAI has released a new, less-restricted AI model aimed at helping cyber defenders, likely enhancing security operations with fewer limitations.

Eric Wallace announces the release of GPT-5.6-Cyber, a model focused on cybersecurity tasks like exploit development, and urges Sam Altman to consider using it for defensive purposes.

A user expresses excitement about a new AI model release and speculates whether Qwen 3.8 27B can compete with Muse Glimmer 30B.

Meta announced it will open source its Muse Spark 1.2 and Muse Glimmer 30B models, described as the biggest open weights since Llama 4 and 3.

Meta announces a new open-source model, Muse Glimmer 30B, released under the Apache-2.0 license.

Meta announces a new open-source model optimized for on-device deployment, aiming to bring efficient AI inference to edge devices.

According to The Wall Street Journal, OpenAI, which once sparked the AI boom, has now fallen behind rival Anthropic and is working hard to catch up. Recently, OpenAI released a super app and a new model, while also benefiting from the industry's backlash against Anthropic.

Guillermo Rauch highlights Grok Imagine Image 2.0, now available on Vercel AI Gateway and ranking #2 on Arena.ai's leaderboard. Vercel offers access via AI CLI and a live playground.

Rumors suggest OpenAI's next major model, codenamed 'Doug', will be its largest pre-training yet and make the Fable model look primitive, potentially launching by November.

Grok announces Imagine Image 2.0, a next-generation image model with precision editing, crisp text rendering, improved factuality, and real-world usefulness.

TokenAI, an Egyptian startup, announces Early Access for Horus Cyper Nano 1.0 BETA, a specialized cybersecurity model for offensive security and red teaming research, with open weights planned for September 2026.

Qwen releases FP8-quantized weights for Qwen3.8-2.4T-A95B, a 2.4T-parameter MoE model with 95B activated parameters, claiming Qwen-Max-class capability in an open release with strong coding, agentic, and long-context performance.

Sam Altman announces that the Astra model is powerful and OpenAI is working to make it generally available, while taking extra time to ensure safety given its cyber capabilities.

Seedance 2.5 launches on Lovart, offering native 30-second 4K video output and up to 50 reference assets for improved consistency, which could be a significant step for AI video generation.

Unsloth releases new GGUF quantizations of Kimi K3, ranging from 466GB to 649GB, enabling efficient deployment of the large model.

Release of Ling-3.0-tiny, a hybrid reasoning model with 7.9B total parameters and only 1.3B active per token, free for a week.

Kimi K3, an open-weight model from Kimi Moonshot, is now generally available in GitHub Copilot, offering frontier-level agentic coding abilities at cost-effective pricing.

Announces the open release of Qwen3.8-2.4T-A95B (aka Qwen3.8-Max) next Wednesday, with a ModelScope link.

Meta released Muse Spark 1.2, a coding-focused update with improvements in code generation, debugging, and codebase understanding, co-trained with the new Muse Code coding agent. The model offers a discounted 'contributor' pricing tier for users who allow Meta to use their data.

A new AI model called Muse Spark 1.2 is announced as essentially free to use, though there is a catch involved.