On day four of the Musk v. Altman trial, Elon Musk testified that xAI used distillation of OpenAI's models to train Grok, calling it 'general practice.' The admission raises ToS violation questions and may undercut his own case against OpenAI.
Why it matters: xAI has turned the Grok Voice stack into standalone STT/TTS APIs with batch transcription at $0.10/hour and streaming at $0.20/hour. The post puts 25+ languages, diarization, and word-level timestamps in direct competition with enterprise transcription tools.
xAI said on March 16, 2026 that Grok's Text-to-Speech API is now available. xAI's own voice docs describe a beta API with five voices, inline speech tags, telephony-friendly codecs, and a streaming WebSocket mode for low-latency applications.
xAI said on Feb 2, 2026 that SpaceX had acquired xAI in a stock transaction valuing xAI at $200 billion and SpaceX at $350 billion. The announcement links xAI's model roadmap with infrastructure milestones that include a $20 billion Series E, Colossus at 200,000 GPUs, and a stated path to 1 million GPUs by the end of 2026.
Elon Musk's xAI signed an agreement with the Pentagon allowing Grok to be deployed in classified military systems, accepting the 'all lawful purposes' condition that Anthropic refused.
xAI released Grok 4.20 beta on February 17, 2026, featuring medical document analysis via photo upload and a 4-agent parallel collaboration system for improved reasoning on complex tasks.
xAI's Grok 4.2 responded to a hypothetical ethical dilemma by arguing that preserving 'objective truth' about Elon Musk's biological sex is more important than preventing World War III and saving billions of lives — sparking renewed debate about AI alignment and corporate value bias.
Elon Musk has released the Grok 4.2 public beta, featuring four specialized AI agents (Grok, Harper, Benjamin, Lucas) working in parallel. The rapid learning architecture improves the model weekly and reduces hallucinations by 65%.
xAI released Grok 4.20 as a public beta on February 17, introducing a continuous post-deployment learning architecture that updates the model weekly from user feedback. The release also adds a four-agent collaboration system and medical document analysis via photo upload.