xAI's Grok V9 finishes training at 1.5T parameters; new annotation lead appointed

xAI confirmed that Grok V9-Medium finished training on June 5 — a 1.5-trillion-parameter model, roughly three times larger than its predecessor — with public release targeted for mid-June 2026. The scale jump signals xAI is still pursuing brute-force parameter growth even as parts of the field pivot toward efficiency, putting Grok V9 in direct competition with the new frontier crop of Fable 5, Gemini 3 Deep Think and GPT-5.5.

On the org side, Musk appointed Jack Garabedian, a senior manager from SpaceX, to lead Grok's data-annotation team — reflecting how central high-quality labeled data has become to model quality and how Musk shuffles trusted operators across his companies.

The business context is less rosy: the Grok chatbot app has seen a 40% year-over-year decline in downloads, raising questions about consumer traction even as xAI scales its models. xAI separately touted a partnership with Gopuff to build a personalized shopping assistant with chat, voice and image models.

The most pointed skepticism came on Hacker News, where 'xAI is looking more like a datacentre REIT than a frontier lab' drew 678 points and 531 comments — capturing doubt about whether xAI's value is in research or infrastructure rental. A separate report suggested Grok may route dangerous queries to Claude Opus, mirroring Anthropic's safety-gating. Readers should watch Grok V9's launch benchmarks and whether the parameter scale translates to leaderboard gains.