AI Launch Radar

• 1.6T-parameter open-source model rivals Opus and GPT-5 while activating just 49B params via 256-expert MoE (9 active per token)
• Trained on Huawei...

• Lowfat is an open-source CLI that filters tokens between you and LLMs, cutting usage 91.8% while keeping the same answers at roughly 1/10th the...

EvoDS tackles static skills and context limits in LLM agents with Autonomous Skill Acquisition and Adaptive Context Compression, enabling...

New Model Releases

• 🔥 Gemma 4 12B: Google released Gemma 4 12B, an open encoder-free multimodal model supporting text, image, and native audio...

• Apple partners with Google Gemini to rebuild Siri, adding on-screen awareness and cross-app data access to compete with Samsung Galaxy AI.
• New...

Superfact solves the friction of exporting real work from AI chats by letting users type one command like "superfact me into a summary" in Claude,...

Microsoft positioned itself as an AI agents company at Build 2026 with three major moves reshaping enterprise tooling.

• Project Solara extends...

Zapier open-sourced its GTM agents as a GitHub repo of agent skills and pre-built automations, built on its platform.

Enterprise AI is moving away from flat-rate subscriptions toward usage-based pricing that mirrors cloud compute bills.

• Agentic tools break the...

Claude now writes 80%+ of merged production code at Anthropic, driving 8x daily merges versus 2024.

• Engineers report 52x speedup on optimization...

Google's Gemma 4 12B delivers near-26B performance in a compact package that runs locally on 16GB RAM devices.

• Encoder-free design processes images...

MMPO introduces Belief Entropy to deliver fine-grained supervision for memory policies in LLM agents, explicitly penalizing summaries that increase...

Glean now supports NVIDIA Nemotron 3 Ultra, giving enterprises an open-weight model option that delivers 91% of frontier LLM performance at lower cost...

Major Open Model Releases

• 🔥 NVIDIA Nemotron 3 Ultra: NVIDIA released Nemotron 3 Ultra, a 550B-parameter open MoE model for long-running agents...

Google's Gemma 4 12B processes images and text in one unified transformer stream, ditching separate vision encoders entirely.

• Consumer access: 12B...

Suno raised $400M to launch a new AI music model developed in direct partnership with the music industry establishment. This approach points to creative tools designed for collaboration with labels and rights holders rather than conflict.

A wave of capable open-weight releases now runs fully on single consumer GPUs, cutting cloud dependency for local AI builds.

• Qwen 3.6-27B delivers...

• Copilot autopilots launch with Scout agent for autonomous desktop tasks
• New MAI models benchmarked against frontier systems for local...

NVIDIA's Nemotron 3 Ultra targets the shift from chatbots to long-running agents that plan, use tools, code, and maintain context across complex...

Enterprises must abandon single-cloud strategies for connected ecosystems as AI workloads spread across clouds, data centers, and edge, making...