minimax m3: sparse attention hits 15.6x decoding speedup at 1m tokens
hermes cut startup time 63% and just beat codex on speed
grok foundation model v9-medium training completed – looks like it's grok 4.5
ai’s biggest shift this week had nothing to do with smarter models
after a few weeks, memory-powered agentic ai strikes again in github trends
deepseek decision makes it the cheapest among the smart models
microsoft could be spending up to ~$233m/mo on claude tokens
telegram secretary bots: your dms on autopilot
google vs apple: building ai phone in 2026
meta laid off 8,000 today. that pushes 2026's ai layoff tally past 75,000 across 15 big companies