Why it matters
Early access is the new release signal. Before a model card or public API exists, where a lab chooses to deploy its next system tells you what it is optimizing for. xAI putting Grok 4.5 inside SpaceX and Tesla first frames the model around engineering and code-heavy internal workflows rather than consumer chat, and it keeps the competitive comparison squarely on Anthropic and OpenAI's frontier tier.
What was announced
Elon Musk said on June 28, 2026 that Grok 4.5 has entered private beta at SpaceX and Tesla, the first deployment of the model ahead of any wider release. According to Musk, Grok 4.5 is built on xAI's 1.5 trillion-parameter V9 foundation model, which completed training on May 26, 2026, and uses data from the Cursor coding environment in supplemental training. Musk said early evaluations place its performance close to, and potentially above, Anthropic's Claude Opus.
Practitioner read
Treat the performance claim as a vendor assertion until xAI publishes benchmarks or a system card; there is no independent evaluation, no eval suite, and no methodology attached to the Opus comparison. The Cursor-data detail, as described by Musk, points to a model tuned for technical reasoning and code generation, which aligns with the engineering needs at both companies hosting the beta. Restricting first access to Musk-controlled firms also gives xAI a controlled, high-volume real-world testbed before exposing the model to outside red-teaming.
What to watch
- •Whether xAI ships a public Grok 4.5 with reproducible benchmarks and a system card, or keeps it internal.
- •Independent coding and reasoning evals that test the near-Opus claim.
- •The stated cadence of new models trained from scratch, which would pressure rivals on release frequency rather than only on peak capability.
- •Colossus 2 is running seven concurrent training jobs including Grok 5 variants at 6T and 10T parameters - if the 4.x models land on schedule, the Grok 5 timeline becomes credible infrastructure evidence, not just a Musk claim.
Key Points
- 1Musk says Grok 4.5 has entered private beta at SpaceX and Tesla, its first deployment ahead of any public release.
- 2The model is built on xAI's 1.5 trillion-parameter V9 foundation and adds Cursor coding data in supplemental training, per Musk.
- 3Internal dogfooding signals an accelerating xAI release cadence, but no benchmarks or system card back the near-Opus performance claim yet.
Scoring Rationale
A private beta of an unreleased frontier model at two Musk-controlled engineering firms is a meaningful competitive signal, but the near-Opus performance claim is vendor-asserted and unverified by independent evals. xAI's Colossus 2 running seven concurrent training runs (including Grok 5 at 10T parameters) adds real infrastructure context and raises the story's stakes for practitioners tracking lab compute trajectories. Score held at 6.6 - no shipped product, no benchmarks, but the Colossus 2 scale and monthly cadence confirm xAI is mounting a credible structural challenge to OpenAI and Anthropic.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
