xAI preps 1.5T-parameter Grok 4.5 in private beta, 2T model expected in August

Built on the V9 foundation, Grok 4.5 at 1.5 trillion parameters represents roughly a threefold scale-up over the existing Grok 4.x series, and it's already in private beta inside Musk's own companies, SpaceX and Tesla, leveraging their data flywheels. A 2-trillion-parameter successor is reportedly slated to complete training by late July for an August public launch.

The headline claim — that early internal evaluations put Grok 4.5 at parity with or above Claude Opus — is exactly the kind of vendor-sourced benchmark that warrants caution until independent testing confirms it. Internal SpaceX/Tesla deployment gives xAI real-world usage data but also means the beta is not yet externally validated.

The timing puts xAI in direct competition with the week's flurry of frontier releases: Anthropic's Sonnet 5, OpenAI's GPT-5.6 Sol, and Google's Gemini updates. The community is watching whether the 2T model challenges the top tier and what it implies for compute demand — a 2T-parameter model is an enormous training and inference commitment, feeding into the same GPU-scarcity dynamics driving AWS price hikes and NVIDIA's revenue-share deals. Watch for external benchmark results and whether the August public launch slips.