model-release

Atlas is introduced as a foundation model for spatial intelligence in a video that features atmospheric music and sparse spoken words.

OpenAI announced the upcoming release of its Astra model, which is the first to achieve 'critical' cyber capabilities, allowing it to find and exploit software vulnerabilities. The company plans a public release soon, with advanced features limited to select partners for safety reasons.

A tier list is provided for the new Fable 5.1 model, recommending Low for UI/UX and mechanical work, High for complex and reasoning tasks, and Max for final checks on large PR work.

Anthropic introduces Claude Fable 5.1 and Claude Mythos 5.1, described as the world's most advanced models for coding and knowledge work.

New Gemma AI models are being featured on the arena AI platform, with speculation about a possible Gemma 5 or related version.

LeVJEPA introduces a more efficient self-supervised video pretraining approach by eliminating the need for target networks and predictors, using a single shared encoder with SIGReg regularizer, and achieving competitive performance with lower compute costs.

dealignai released a cybersecurity-focused weight-modified variant of the 753B-parameter GLM-5.3-FP8 model that reduces refusals for offensive-security tasks such as red-teaming, exploit development, and malware analysis while preserving base capabilities and running natively on Hopper GPUs via stock vLLM.

The article advocates for using a strict pass metric instead of partial credit to evaluate finance AI agents, referencing the Ling-3.0-flash-Fin release to improve reliability in complex, step-dependent tasks.

The article discusses the benchmark card for Ling-3.0-flash-Fin, highlighting that the results are based on specific agent systems and evaluation pipelines rather than raw model performance.

Dealignai has released GLM-5.3 Uncensored NVFP4, an uncensored AI model designed for offensive cybersecurity tasks with refusal behavior removed at the weight level, capable of generating working security code while maintaining near-base performance.

This article presents official 1-bit quantization builds for the Hy4 preview model, offering GGUF files with reduced sizes and instructions for running on patched llama.cpp.

Tencent AI has announced the Hy4 preview model with 770B total parameters, 49B active parameters, 1M context, and open weights, marking a significant update in the competitive Chinese AI landscape.

This repository provides GGUF quantizations of the Qwen3.8-27B AI model using GSQ and RCO methods for efficient deployment in standard tools.

Ant's Ling team has released Ling-3.0-flash-Fin, a finance-enhanced AI model designed for financial workflows, with free access on OpenRouter for one month and plans to open-source the weights.

HuggingModels announces Vexion GPT Medium, a compact AI model trained on CulturaX for fluent Russian language and efficient text generation, targeting developers needing fast localized NLP solutions.

Tavus introduces Sparrow-2, a state-of-the-art real-time conversational understanding AI model that helps voice AI systems decide when to listen, wait, speak, or keep speaking during conversations.

GLM-5.3 AI model weights are set to be released on August 28, 2026, on Hugging Face, featuring frontier coding and emergent cyber capabilities.

The article announces that GLM-5.3, an AI language model, is coming soon.

Z.ai has officially released GLM-5.3-Flash, a natively multimodal AI model with 320B parameters and a 1M-token context window, previously previewed as Ox Alpha, available via Modular Cloud and running on Chinese AI chips under an MIT License.

Gemini 3.5 Transcribe is announced as the most precise speech-to-text model yet, with discussion links provided on Product Hunt.