SpaceXAI released Grok 4.5 on July 8, 2026, its first model launch since the company went public several weeks earlier. Elon Musk called it an "Opus-class model" in a post on X, and said SpaceXAI's internal benchmarking found it "roughly comparable to Opus 4.7, but much faster" TechCrunch, x.com/elonmusk. Public availability is set for July 9.

The model is priced at $2 per million input tokens and $6 per million output tokens, against $5 and $25 respectively for Anthropic's Opus 4.7 TechCrunch. SpaceXAI is marketing this alongside a claim of "twice greater token efficiency" than competing frontier models, a phrase that appears in both the TechCrunch report and xAI's own blog post. Musk framed the release timing as a response to what he described as strong positive feedback from customers in SpaceXAI's beta program, rather than a response to any single competitor's roadmap.

The launch lands one day ahead of OpenAI's planned release of GPT 5.6, scheduled for July 9 according to reporting from Politico and confirmed in TechCrunch's coverage. Whether the sequencing was deliberate is not something either company has stated on the record; SpaceXAI's own framing, per Musk's posts, centers on beta feedback rather than competitive scheduling.

Architecturally, Grok 4.5 is built on a 1.5 trillion-parameter "V9" foundation model, with Cursor data folded in during supplemental training, according to a Musk post from June 28 — ten days ahead of the public unveiling x.com/elonmusk. The model ran in private beta inside SpaceX and Tesla before the wider release, giving SpaceXAI internal production workloads as a testing ground well before external customers saw it. That internal-dogfooding approach mirrors how several frontier labs have validated coding and agentic capability under real operational load rather than benchmark suites alone.

The Cursor relationship is now formalized as a partnership, not just a training-data arrangement. Coverage from TheNextWeb, Moneycontrol, and Investing.com describes Grok 4.5 as jointly positioned with Cursor for coding workflows, alongside a stated push into legal and finance tasks — domains where document-heavy, high-context-window reasoning and tool use matter more than raw chat fluency. SpaceXAI's own materials describe the model as its smartest yet, built specifically for coding, agentic tasks, and knowledge work, and the company is offering a free tier or trial at launch to drive adoption xAI.

The release follows a $6 billion Series C round for xAI, whose investor list is notable for its breadth: A16Z, Sequoia Capital, Lightspeed, and Valor Equity Partners sit alongside sovereign and institutional capital including MGX, QIA, OIA, Kingdom Holdings, BlackRock, Fidelity Management & Research Company, and Morgan Stanley xAI blog. That mix of Silicon Valley venture money and sovereign wealth funds is now fairly standard for frontier AI labs raising at this scale, but the presence of both BlackRock and Fidelity as named investors signals institutional asset managers treating foundation-model equity as a distinct allocation rather than a speculative side bet.

X, the social platform, remains a subsidiary of SpaceXAI, a structural detail that matters for distribution: Grok models ship with a built-in consumer surface that OpenAI and Anthropic must otherwise pay to reach through third-party integrations or their own standalone apps.

On pricing, the gap is stark enough to warrant scrutiny beyond the topline numbers. At $2/$6 per million tokens versus Opus 4.7's $5/$25, SpaceXAI is undercutting on output-token cost by roughly a factor of four, which matters disproportionately for agentic workloads that generate long chains of tool calls and intermediate reasoning tokens rather than short chat responses. Whether that price gap survives sustained inference demand at scale, or gets adjusted once free-tier usage settles into paid conversion, is not yet knowable from Monday's announcement alone.

The broader context here is a market where "Opus-class" has become an informal shorthand for a capability tier rather than a specific vendor's product, similar to how "GPT-4-class" briefly served as an industry benchmark two years ago. That a competitor is now defining itself relative to Anthropic's naming, rather than OpenAI's, says something about how the competitive map among frontier labs has shifted over the past year. Coding and agentic benchmarks, not general chat quality, appear to be where labs believe the next round of enterprise contracts will be won or lost, and SpaceXAI's Cursor partnership and finance/legal positioning both point in that direction.

In this author's view, the token-efficiency claim deserves the same skepticism any vendor-issued benchmark deserves until independent evaluators — LMSYS, Epoch AI, or enterprise procurement teams running their own workloads — reproduce it under comparable conditions. Self-reported efficiency and cost figures have a long history of not surviving contact with real-world usage patterns, and this is a pattern worth watching rather than taking at face value.