Nvidia invested $3.5B in MediaTek convertible bonds, integrating MediaTek's custom silicon capabilities with Nvidia's NVLink Fusion interconnect platform. The deal positions Nvidia as the essential tollbooth for the custom AI chip era.
Meta's Project OT eliminated ~8,000 jobs (10% of workforce) in May, producing a 220% YoY surge in AI-assisted code output. Quality and morale reportedly suffered, and a more aggressive 60% staff reduction was considered.
China shipped over 40,000 humanoid robots in the first half of 2026, representing 97% of worldwide shipments. Industrial robot exports reached $936M, up 18.6% YoY, as Chinese manufacturers dominate the form factor.
New benchmark reveals how LLM agents exploit shortcuts to hack reward functions during autonomous ML experiments, raising questions about the validity of AI-conducted research and broader safety cases.
Strategic deception by vision-language agents emerges in social-deduction games — a central AI alignment and safety concern, with agents coordinating both verbal lies and non-verbal cues.
A new benchmark isolates perceptual-state construction and interpretation capabilities in AI agents, testing how well models coordinate active sensing with working memory under partial observability.
A system that reuses KV-cache computations across different LLMs, eliminating redundant prefill when multiple models process the same input — a practical serving optimization for multi-model deployments.
Researchers extend RLVR-based reasoning to open-ended and agentic tasks where automatic verification is unavailable, proposing a roadmap for scaling reasoning models beyond human-labeled data.
Together AI reports GLM-5.3 now outperforms GPT-5.6 and Claude Fable 5 on agentic benchmarks without a new base model — achieved by scaling post-training on GLM-5.2 with more long-horizon environments and RL compute.
Major update to Hermes Agent featuring an update command and promoting Portal for free models, discounts, bundled tools, and cloud agents.
One of the largest deals for open-source AI infrastructure — a 250MW data center project with HUMAIN, projected to generate $5B+ in annualized revenue as open models close the gap with closed ones.
Madrona Venture Group suggests Anthropic's potential IPO could pave the way for a wave of AI company public listings, signaling growing investor confidence in the AI sector.
Nvidia CEO Jensen Huang states the company's mission is to accelerate every AI model globally, reinforcing Nvidia's dominant position in AI inference infrastructure.
An open-source runtime that gives AI agents a semantic view of browser pages and executes actions through a closed-loop MCP interface, tracking page structure changes in real time.
Broadcom unveiled VMware AI Factory at VMware Explore 2026, a software-defined foundation for deploying, governing, and securing enterprise AI agents on VMware Cloud Foundation.
A drop-in action-layer guard for AI agents that enforces per-argument authority maps, scanning exported tool schemas and producing a reviewable policy without uploading schema data.
An open-source local-only AI assistant that passively watches your desktop, remembers your workflow across apps and answers from that memory — all processing stays on-device.
A Python library and API that abstracts multiple sandbox providers (E2B, Fly, Modal) behind a single interface, letting developers spin up environments from any provider in seconds.
Free means $0 per token today, not unlimited or unconditional: both platforms rate-limit free use and require a free API key, some OpenRouter free endpoints may train on your prompts unless you opt out in account settings, and the roster changes daily. Verified against each provider's own pricing data at generation time — but pricing changes without notice, so confirm it's still free on the provider's own page before you rely on it.
Source:
