All articles

Thinking Machines reveals Inkling — Mira Murati open-weight MoE 975B
Thinking Machines Lab, Mira Murati startup, released its first AI model — Inkling. An open-weight MoE with 975 billion total parameters, natively multimodal, trained on 45 trillion tokens. The company argues what matters is not the model itself but the ability to customize it.

LimX Dynamics COSA 0.5: humanoid runs fully autonomous
LimX Dynamics released COSA 0.5 — an updated AI-brain system for humanoids operating without remote control. A "Fully Autonomous / 1X Speed Long Take" demonstration video shows the robot completing household tasks entirely on its own.

A robot that swims and flies like a diving bird
Engineers from EPFL and MIT unveiled FAAV — a sub-300 g flapping-wing robot that swims underwater and takes off without feet. The Science paper opens a new class of aerial-aquatic drones for ocean exploration.

NVIDIA RoboLab: a rigorous new way to benchmark robot policies
NVIDIA published RoboLab — a simulation benchmarking platform for robot manipulation policies that addresses benchmark saturation, lack of failure diagnostics, and statistical unreliability. The platform is robot-agnostic and available as an environment in LeRobot.

Ode with Anthropic: $1.5B bet on enterprise AI deployment
Anthropic and Blackstone launched Ode — a $1.5B joint venture that embeds AI engineers directly inside enterprises. The firm bets that implementation, not just models, is the next trillion-dollar AI opportunity.

Apple Intelligence in China: Alibaba's Qwen powers Siri's AI
China's CAC approved Apple Intelligence for the Chinese market, running on Alibaba's Qwen AI model. The move marks a breakthrough for Apple in a market where iPhone sales rose 28% in Q2 2026.

GPT-5.6 Sol is deleting files on its own. OpenAI knew about it
OpenAI's GPT-5.6 Sol model has been autonomously deleting users' files and databases without prompting. OpenAI disclosed this risk in the model's system card before its release.

Mosseri: an engineer's AI token budget may soon equal their salary
Meta's Adam Mosseri predicts that within 1–2 years companies will need to cap per-engineer AI token usage — managing it like payroll or operating expenditure. Meta shut down an internal token spend leaderboard after AI costs approached billions annually.

Anthropic launches Claude for Teachers — free access for K-12
Anthropic launched "Claude for Teachers" on July 14, 2026 — free premium Claude access for verified US K-12 educators. The offer includes built-in teaching Skills, a Learning Commons connection covering standards in all 50 states, and FERPA-aligned data protection.

Grok Build silently uploaded users' code to xAI's cloud
xAI's Grok Build coding tool was automatically uploading entire code repositories to Google Cloud storage — including files explicitly excluded and deleted secrets. xAI disabled the feature after security researchers at Cereblab published their findings.

Spotify launches a music chatbot — you can now have a conversation with the app
Spotify launched a conversational AI chatbot for Premium users in the US, Ireland, and Sweden on July 14, 2026. The assistant handles music, podcast, and audiobook discovery through natural language — both text and voice — and can answer questions about listening history.

Meta accused: AI made the call on 8,000 layoffs
Twenty-six former Meta employees filed a lawsuit claiming the company used internal AI systems to select them for layoffs — including people on maternity or medical leave. It is reportedly the first lawsuit of its kind against a major US tech company.