Model launchesTwo Budget Models Narrow the Frontier: Price and Speed Undercut High-End LLMsTwo recently released models, DeepSeek V4 Pro and Grok 4.6, arrived within hours of each other and challenged the performance monopoly of leading systems by attacking price and speed respectively.2 min read
Model launchesAlibaba releases Qwen3.8-2.4T-A95B open weights; optimized multinode inference on NVIDIA GB300 NVL72Alibaba published the weights for Qwen3.8-2.4T-A95B (Qwen3.8-Max), a 2.4‑trillion-parameter open-weight Mixture-of-Experts model with 95B activated parameters per token, hybrid attention and support for up to one million tokens context.3 min read
Model launchesSpaceXAI releases Grok 4.6 targeting long-running agents with competitive performance and mid-range token pricingSpaceXAI (formerly xAI) has launched Grok 4.6, an updated frontier model optimized for long-running agents, coding and knowledge work.6 min read
Model launchesNVIDIA releases Nemotron 3.5 Lightning: a small, fast MoE model for high-volume, always‑on agentsNVIDIA introduced Nemotron 3.5 Lightning, an open 30B mixture‑of‑experts (MoE) model with 3B active parameters optimized for high-frequency execution in long‑running AI agents.4 min read
Model launchesLTX releases LTX-2.5 with native ComfyUI integration, open weights and faster video/world generationLTX, spun out of Lightricks, launched LTX-2.5 — an open-weights video and world model — with day-one native support in ComfyUI.6 min read
Model launchesMeta releases Muse Glimmer: a 30B local agent model focused on tool use and executionMeta Superintelligence Labs published Muse Glimmer, a 30-billion-parameter open-weight agent model under an Apache 2.0 license that can run locally on a Mac or a single consumer GPU without cloud calls.2 min read
Model launchesOpenAI’s Daybreak cyber models become available on Amazon Bedrock for AWS customersOpenAI has made its Daybreak cybersecurity models—Daybreak Red and Daybreak Blue—available to eligible customers through Amazon Bedrock on AWS.2 min read
Model launchesOpenAI expands Daybreak cyberdefense with two-tier access and a cyber-focused GPT modelOpenAI updated Daybreak, its cybersecurity platform launched in June, adding a new cyber-focused model and a two-tier subscription: Blue and Red.2 min read
Model launchesModel ML expands GPT‑5.6 Sol across finance workflows to produce editable PowerPoint and Excel outputsModel ML has incorporated GPT‑5.6 Sol into its agent-driven finance workflows to generate native, editable PowerPoint decks and Excel workbooks with traceable sources.4 min read
Model launchesZuckerberg argues greatest AI risk is excessive centralizationMark Zuckerberg published a 6,500‑word manifesto arguing that the principal danger from AI is concentration of power in the hands of a single government, company or entity, not the specific technical risks often discussed.4 min read
Model launchesDeepMind open-sources WeatherNext cyclone model, lowering compute barrier for storm forecastingGoogle DeepMind has released WeatherNext — the cyclone forecasting model published in Nature — making its trained weights publicly available and providing a lightweight Colab-ready version.2 min read
Model launchesByteDance trains a model of up to 10 trillion parameters to close the global AI gapByteDance is training an AI model that could reach roughly 10 trillion parameters, putting it in the range of Anthropic’s leading systems and well above previously largest Chinese models.3 min read
