Elon Musk announced Wednesday that SpaceXAI will make Grok 4.5 publicly available Thursday, July 9, posting the news on X. Musk pitched the model as Opus-class, a direct comparison to Anthropic’s flagship Claude Opus line, and said it beats Opus on speed, token efficiency, and price. Bloomberg reported the announcement the same day Musk posted it.
The model runs on a foundation xAI calls V9, which the company says holds 1.5 trillion parameters. Musk did not publish benchmark results alongside that figure. xAI also has not detailed how those parameters are activated per query, a distinction that matters directly for the cost and latency claims Musk is making.
SpaceXAI is the renamed xAI unit that Musk folded into SpaceX earlier this year, a restructuring that put Grok’s development under a company that now trades publicly as SPCX. Thursday’s release follows Grok 4.3, which shipped in April, and a private beta run internally across SpaceX and Tesla over the summer. Testers on that beta gave positive marks, per Bloomberg, though the outlet did not cite specific metrics or name individual reviewers.
Separate reporting ties Grok 4.5 to Cursor, the AI coding startup, describing it as the first model the two companies built together. That collaboration positions Grok 4.5 as a coding rival to Opus 4.8 and GPT-5.5, the two models most coding tools default to today. Distribution at launch runs through Grok on X and the standalone Grok apps; SpaceXAI has not published a rollout schedule or said whether Grok 4.5 will require a new subscription tier.
Musk’s claims about speed, token efficiency, and cost are his own. Neither Bloomberg’s report nor Musk’s post includes independent benchmark results, and SpaceXAI has not published a technical card or third-party evaluation alongside the launch. xAI’s model launches typically precede outside testing by days or weeks, and Grok 4.3’s April release followed that same sequence.
The timing puts Grok 4.5 in the middle of a stacked week at the top of the frontier market. Anthropic brought Mythos and Fable back online this week and released Sonnet 5. OpenAI, meanwhile, plans to launch three GPT-5.6 variants, Sol, Terra, and Luna, this week after holding them back at a government agency’s request. Three labs are refreshing their top-tier lineups within days of each other, which will make any single company’s benchmark claims harder to trust without outside verification.
Teams evaluating coding assistants or high-volume inference contracts should wait for independent benchmarks before shifting workloads off Opus or GPT-5.5 on Musk’s word alone. The more concrete signal to watch is whether Cursor ships Grok 4.5 as a selectable model inside its own product. That would confirm the joint-development claim faster than any post on X.
Bloomberg reported the announcement on July 8, 2026, based on Elon Musk’s post on X.
