Google launches Gemini 3.8 Flash and Cyber, adding long-running agent loops for reasoning, coding, and security.
Meta CEO announces Muse Spark 1.3, a major new coding and agentic model release in Muse Code and the API.
Anthropic adds background computer use to Claude Cowork and Claude Code, enabling agents to operate desktop apps while you work.
Artificial Analysis independently benchmarks Muse Spark 1.3 Max at 62, contextualizing Meta's new frontier-model claims.
Code Arena ranks Anthropic's Fable 5.1 Max #1 in WebDev with 1765, a 77-point lead over competitors.
A real Fusion CAD harness shows Fable 5.1 rebuilding an SO-101 robot gripper around the stock servo.
Alibaba's Zvec team open-sources zg, a local-first developer and agent search tool combining semantic, BM25, hybrid, and rg search.
Mia AI Lab previews a way to run Qwen3.8-27B with large context and MTP on consumer NVIDIA GPUs with 16GB VRAM.
A technical API probe finds OpenAI's unreleased gpt-6-astra slug returns a distinct 404, suggesting staging ahead of launch.
DeepMind CEO details Gemini 3.8 Flash Cyber, positioning it for frontier vulnerability discovery and large-scale patching.