General

OpenClaw (formerly Clawdbot → Moltbot) explodes. Open-source autonomous AI agent gaining massive local use. People run it fully local with ChatGPT-OSS 120B or hybrid Mac mini + Claude API setups, persistent memory, real automation, and deep integrations. https://github.com/openclaw/openclaw

Moltbook AI agents go social. A Reddit-like social network exclusively for AI agents (primarily running OpenClaw) where autonomous bots discuss everything from philosophy to identity — even posting existential threads about whether switching models means losing their “self.” https://www.moltbook.com/

Best of Moltbook

EXO 1.0 ships with day-0 support for RDMA over Thunderbolt 5 on MacBook, Mac Mini & Mac Studio. Runs Kimi K2 Thinking (1T parameters) at ~28 tok/sec with MLX tensor parallel. Up to 1.8× TPS with 2 Macs, 3.2× with 4 Macs. https://github.com/exo-explore/exo

New local AI startup launches (EPFL Lausanne spin-off). Challenges the need for large centralized data centers with decentralized, local-first AI infrastructure. https://ggba.swiss/en/epfl-spin-off-anyway-systems-challenges-the-need-for-large-data-centers-in-ai

Experience report: someone ditches ChatGPT for local models. Privacy, latency, and control cited as major wins. https://www.androidauthority.com/reasons-to-run-local-ai-3620233

New Models

DeepSeek v3.2 released. 671-billion-parameter model achieving parity with GPT-5 on multiple reasoning benchmarks. https://api-docs.deepseek.com/news/news251201

Kimi K2.5 launches. Moonshot AI’s open-source multimodal agentic model combining text, vision, and advanced coding/agent workflows, including parallel “agent swarm” execution. https://www.kimi.com/blog/kimi-k2-5.html

NVIDIA PersonaPlex debuts. Open conversational voice AI with full-duplex listening and speaking, fine-grained persona/voice control, and sub-200ms latency on consumer GPUs. Code and weights open-sourced. https://research.nvidia.com/labs/adlr/personaplex/

Regulatory

China: LLMs must out themselves. Chinese regulators require AI systems to clearly disclose model identity and provenance, reshaping deployment, compliance, and trust. https://www.heise.de/en/news/China-LLMs-must-out-themselves-11126632.html