Model Releases · JUNE 28, 2026
Grok 4.5 ships to SpaceX and Tesla only: 1.5T parameters, V9 foundation, Cursor-tuned
xAI's V9-based Grok 4.5 entered closed beta on June 28 inside Musk's two engineering companies — 50% larger than Grok 4.4, trained with Cursor data, and benchmarked internally against Claude Opus.
Elon Musk announced Grok 4.5 on X on June 28, but the model isn't going anywhere a competitor's evaluator can reach it. xAI shipped the 1.5-trillion-parameter system directly into closed beta at SpaceX and Tesla, skipping the public benchmark cycle entirely and routing the model into the only two consumer-facing companies Musk also runs.
The architectural specs come via Crypto Briefing's writeup. Grok 4.5 sits on a new V9 foundation, roughly 3x the size of the V8-small architecture that underpinned earlier Grok 4 variants, and 50% larger than the 1-trillion-parameter Grok 4.4 that landed in late May. The May timeline had originally pegged 4.5 for the same window; it slipped about a month.
The captive-beta structure matters more than the parameter count. Real-world use inside SpaceX and Tesla feeds telemetry back into the model alongside reinforcement-learning work and an internal tooling layer Musk referred to as Grok Build. Training data reportedly includes signal from Cursor, the AI-assisted code editor. Early internal evaluations, described to Crypto Briefing, characterize the model's performance as "comparable to, if not superior to" Anthropic's Claude Opus.
That's a claim, not a result. Without a public eval, it functions as narrative management aimed at the audience that matters for xAI's next funding round.
The economics behind the cadence are visible in the SpaceX S-1 filed in May 2026. xAI lost $6.4 billion in 2025 on $3.2 billion in revenue, up from a $1.56 billion loss on $2.62 billion the year before. AI-segment capital expenditure ran $12.7 billion across 2025 and another $7.7 billion in Q1 2026 alone, an annualized run rate near $30.8 billion. The Colossus datacenter came online in 122 days, Colossus II in 91, together delivering roughly a gigawatt of compute. Grok 4 itself was trained on a 200,000-GPU Colossus cluster and launched in July 2025 with a 256,000-token context window and a $300-per-month SuperGrok Heavy tier.
The S-1 also previews the trajectory, describing the next generation of Grok as scaling to "multiple trillions of parameters" and framing the jump as a "step change in reasoning in depth and overall intelligence." Crypto Briefing's reporting on the roadmap puts numbers on it: Grok 5 variants are projected at 10 trillion parameters, and xAI has committed to releasing models trained from scratch at SpaceX every month for the rest of 2026.
The release pattern is the story. A frontier lab burning capex at a $30-billion run rate has decided that proprietary engineering telemetry from two affiliated companies is a better training signal, and a better marketing surface, than the public leaderboard culture that defined the 2023–2024 model cycle. The benchmark era is being privatized inside Musk's own org chart.
Sources
- https://cryptobriefing.com/grok-4-5-private-beta-spacex-tesla/
- https://cryptobriefing.com/xai-grok-4-5-v9-model-upgrade/
- https://techcrunch.com/2026/05/20/xai-burned-6-4b-last-year-spacexs-ipo-filing-shows-why-the-spending-is-far-from-over/
- https://techcrunch.com/2025/07/09/elon-musks-xai-launches-grok-4-alongside-a-300-monthly-subscription/
- https://x.ai/news/grok-4