Grok 4.5 Enters Private Beta at SpaceX and Tesla — And Musk Says It’s Already Rivaling Opus

Grok 4.5 Enters Private Beta at SpaceX and Tesla: Elon Musk just dropped one of his more eyebrow-raising AI claims yet — and this time it’s not about AGI timelines. In a post on X, Musk confirmed that Grok 4.5, built on xAI’s massive new foundation model, is quietly running inside two of his own companies.

What Musk Actually Said

Musk confirmed Grok 4.5 is built on xAI’s V9 foundation model, a 1.5-trillion-parameter system, with Cursor data added during supplemental training, and it’s now deployed in private beta at SpaceX and Tesla. He noted that early evaluations put its performance close to — and potentially exceeding — Anthropic’s Opus.

Why the Cursor Data Matters

The addition of Cursor’s developer workflow data suggests Grok 4.5 has been specifically tuned for technical reasoning and code-related tasks — directly relevant to SpaceX’s engineering pipelines and Tesla’s software development. That’s not a coincidence. This follows SpaceX’s reported acquisition of Cursor, with engineering leads from the coding tool brought directly into xAI to restructure Grok’s coding capabilities from the ground up. We covered the scale of that deal in detail in our SpaceX-Cursor $60 billion acquisition breakdown.

A Pattern of Monthly Releases

Musk’s pledge of a new “completely trained from scratch” model every month is an aggressive cadence, even by xAI’s standards. It builds directly on the V9 foundation model groundwork that finished training in late May 2026 — described at the time as a major leap over the 0.5T v8-small model handling production Grok traffic.

This rapid-fire approach tracks with what TechnoSports has tracked closely across xAI’s recent run, including the Grok 4.1 Fast release that briefly leapfrogged Google’s Gemini 3 on benchmarks.

The “Close to Opus” Claim Deserves Context

It’s worth flagging that this is Musk’s own characterisation based on early internal evaluations, not an independently verified, peer-reviewed benchmark comparison. Independent reporting notes that SWE-bench Verified — the standard coding benchmark used for these comparisons — has grown increasingly contaminated, with researchers and even OpenAI acknowledging that tasks leak into training data, meaning models may partially recall memorised solutions rather than demonstrate genuine engineering skill.

Current SWE-bench leaders Claude Opus 4.6 and Gemini 3.1 Pro sit around 80-81%, while existing Grok 4-series models score closer to 72-75% — so “close to Opus” would represent a genuinely significant leap if it holds up under independent testing.

What This Means for Tesla Owners

If Grok 4.5 graduates from internal beta to production, it could meaningfully upgrade Tesla’s in-car voice assistant — an area that has historically lagged behind xAI’s broader model development pace. Whether that arrives via a standard over-the-air update remains to be seen.

For official confirmation and ongoing updates, keep an eye on xAI’s official website and Musk’s own X account. Given the pace of releases promised here, expect this story to keep moving fast.