Grok 4.5 will be open to the public by Thursday, July 9, according to Elon Musk. Musk stated that the beta program had produced “strong positive feedback” and that the model would ship to the public the next day. 

He called it “an Opus-class model, but faster, more token-efficient and lower cost,” a reference to Anthropic’s Claude Opus tier. However, his post did not include benchmark scores, pricing, or context-window figures.

A rebranded lab and a $60 billion acquisition

xAI has renamed itself SpaceXAI, a change that was announced earlier in the week. Also, the latest model, Grok 4.5, is the first that was built jointly with Cursor, the coding startup that SpaceX (NASDAQ: SPCX) is in the process of buying for $60 billion. 

The model was initially set to launch earlier in the week but was pushed to Wednesday as the team was improving its efficiency, according to The Information, which cited a memo the company sent to staff. However, based on Musk’s post on X, that release may come one day later.

In a June 28 post, Musk gave a technical outline of Grok 4.5, stating, “Grok 4.5, based on our 1.5T V9 foundation model, with Cursor data added in supplemental training, is now in private beta at SpaceX & Tesla.” He wrote that early evaluations “show performance close to, perhaps exceeding Opus.” 

Musk also stated that SpaceX will be releasing new models that are completely trained from scratch every month this year.

A third-party breakdown published by kie.ai, drawing on posts from the leak-tracking account TestingCatalog, put that foundation model at about three times the size of xAI’s existing v8-small, estimated at about 0.5 trillion parameters. 

The report also flagged that the “close to or beyond Opus” line is an internal impression from company evaluators, not a published benchmark. It stated that a larger 2-trillion-parameter model is reportedly still training for an August release.

Is the Grok 4.5 comparison to Opus verified?

The Opus model that Grok 4.5 is being measured against is not yet clear. On the independent Artificial Analysis leaderboard, Claude Fable 5 and Claude Opus 4.8 are the highest-intelligence models available, followed by OpenAI’s GPT-5.5 and Claude Opus 4.7, respectively. 

So, matching that model of Opus, it would place Grok 4.5 near the front of the field.

Jun Song, an AI commentator, predicted on July 8 that Grok 4.5 would land as a 1.5 trillion parameter model performing “slightly worse” than the smaller GLM-5.2 while still drawing heavy promotion on price and openness. 

Although some voices on X are disagreeing with him, given that he has not tested the model.

Stanford’s 2026 AI Index found the leading labs clustered tightly, with Anthropic, xAI, Google, and OpenAI all within 25 Elo points on the Arena leaderboard as of March 2026. With capability converging, the report noted, competition is shifting toward cost, reliability, and speed, the exact axes Musk chose to emphasize.

Independent evaluations after the model is released will show whether Grok 4.5 delivers Opus-level output at a lower price or whether the comparison holds only inside SpaceXAI.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It’s free.