Starcloud added $250M to its March Series A for AI data centers in space, with Nvidia among the backers.
RealMan plans roughly 1,000 RealBOT deployments this year, demoing medicine and food handling at the World Robot Conference.
xAI's Grok 4.6 is now available on Vertex AI for enterprise deployment, with Grok Bot also included in more plans.
A new Weight Cache Daemon slashes weight-loading times to ~0.63s for Ling-2.6-1T FP8 and delivers ~500x speedups on Qwen3-235B via zero-copy CUDA IPC over persistent GPU processes.
An August 26 webinar will cover production serving of Kimi K3, including architecture, quality and optimizations.
Nous shared inference features it has kept in production for months; a thin update with no new model attached.
Schaeffler validated formed strain wave gearboxes for humanoids and will scale production in 2027, cutting manufacturing costs by over 25% and material use by more than 75%.
Galaxy Robot Park opened in Seoul with humanoids performing taekwondo, and hands-on programs for visitors to pilot robot dogs and dance with robots.
A humanoid training for the 2026 World Humanoid Robot Games in Beijing ran too fast during a run event, failed to stop and slammed into a safety cushion in a shower of sparks.
DeepSeek's experimental multimodal variant of V4-Flash adds image understanding at existing token rates, with same-day support in DeepSeek Harness 0.1.1 and the Responses API.
Show HN: OzBrain acts as a 'shared brain' so multiple agents and humans read and write the same trusted context instead of siloed memory.
Show HN: Heimdall is a knowledge layer that hands coding agents verified context so they act on trusted information rather than ungrounded model memory.
Source:
