SpaceXAI released Grok 4.5 on July 8, 2026, its first model launch since the company went public weeks earlier. Elon Musk described it as an "Opus-class model" in a post on X, referring to Anthropic's flagship Opus 4.7. According to Musk and reporting from TechCrunch, SpaceXAI's internal testing found Grok 4.5 "roughly comparable to Opus 4.7, but much faster." The model became publicly available on July 9.

Pricing and Claims

The pricing difference is substantial. Grok 4.5 costs $2 per million input tokens and $6 per million output tokens, compared to $5 and $25 respectively for Opus 4.7 TechCrunch. That's roughly four times cheaper on output tokens — the part of a calculation that gets billed when the model generates longer responses.

SpaceXAI claims the model offers "twice greater token efficiency" than rival models, a metric that measures how much computation gets done per token processed. That claim appears in both TechCrunch's reporting and xAI's own blog post. Musk attributed the release timing to positive feedback from a beta testing program rather than to any competitive move by a rival company — though notably, the announcement came one day before OpenAI's planned release of GPT 5.6, scheduled for July 9 according to Politico and confirmed by TechCrunch. Whether the timing was intentional is unclear; neither company has said on the record.

How It's Built and Where It's Tested

Grok 4.5 runs on a foundation model with 1.5 trillion parameters — think of parameters as the model's internal knobs and weights that determine how it processes information. The model was trained using data from Cursor, a coding assistant, according to a Musk post from June 28 x.com/elonmusk.

Before the public launch, SpaceXAI tested Grok 4.5 in private beta across its own internal operations at SpaceX and Tesla. This real-world testing — called "dogfooding" in the industry — let engineers see how the model performed on actual production tasks rather than just benchmarks. Several other frontier AI labs have adopted the same approach when validating coding and agentic capabilities (where AI systems call tools or perform multi-step tasks) under genuine operational load.

The Cursor Partnership and Target Use Cases

Cursor and SpaceXAI have formalized their relationship into a formal partnership, according to coverage from TheNextWeb, Moneycontrol, and Investing.com. Grok 4.5 is being marketed jointly with Cursor for coding work, and SpaceXAI is also pushing into legal and finance applications — domains where the model needs to reason over long documents and use tools rather than just generate conversation.

SpaceXAI describes Grok 4.5 as its most capable model yet, purpose-built for coding, agentic work, and knowledge tasks. A free tier is available at launch to encourage adoption.

Funding and Market Dynamics

xAI recently closed a $6 billion Series C round with an unusual mix of investors. The list includes venture capital firms like A16Z, Sequoia, and Lightspeed, alongside sovereign wealth funds and institutional money managers such as MGX, QIA, BlackRock, Fidelity, and Morgan Stanley xAI blog. That combination of Silicon Valley venture capital and state-backed wealth funds has become standard for AI labs raising at this scale. The presence of BlackRock and Fidelity as named investors is worth noting: it suggests major asset managers are treating foundation-model equity as a serious allocation, not a speculative bet on the side.

A structural advantage: X, the social platform, is owned by SpaceXAI. Grok models ship with a built-in user interface on X itself, giving them distribution that OpenAI and Anthropic must otherwise buy through third-party partnerships or their own standalone apps.

What This Means for the Market

The term "Opus-class" has become an informal way to describe a tier of capability across the industry, much like "GPT-4-class" served as a rough benchmark two years ago. The fact that a competitor is now positioning itself relative to Anthropic's naming scheme rather than OpenAI's signals a shift in how the AI market sees itself. Coding and agentic tasks — not just chatbot quality — appear to be where frontier labs expect the next wave of enterprise contracts will go, and SpaceXAI's positioning aligns with that outlook.

The gap in output-token pricing deserves closer attention. Since agentic workloads generate much longer sequences of intermediate reasoning than typical chat responses, a four-fold discount on output tokens could make a real difference to customers. Whether that price survives once free-tier users convert to paid usage, or whether demand at scale forces an adjustment, is an open question.

Any vendor-issued efficiency claim ought to be tested against real-world usage before being treated as settled fact. Independent evaluators — benchmarking labs, enterprise customers running their own tests — will eventually tell us whether the efficiency gap holds up in practice. Self-reported metrics have a track record of not surviving contact with actual production workloads, and this is a pattern worth watching rather than accepting at face value.