SpaceXAI releases Grok 4.5, which Elon describes as an ‘Opus-class model’

SpaceXAI has released its latest model, Grok 4.5 — the archetypal since the institution went nationalist respective weeks ago.

In a blog post published Wednesday, SpaceXAI characterized its caller merchandise arsenic a workhorse that tin tackle each of the emblematic tasks that the AI manufacture has sought to automate: coding and app-building, bureau and clerical work, research, writing, and different forms of regular cognition work.

Grok tin supposedly bash each this for little spend, too, arsenic SpaceXAI says that its exemplary has “twice greater token efficiency” than different starring models. If it carries done to real-world usage cases, that ratio would beryllium a large vantage for SpaceXAI, since the outgo of tokens has been a increasing interest for AI consumers.

The institution released benchmark metrics Wednesday that appeared to amusement Grok’s competitiveness with different apical models from SpaceXAI competitors, though conscionable abbreviated of best-in-class:

Image Credits:SpaceXAI

In a post connected his societal media level X (which is simply a subsidiary of SpaceXAI), laminitis Elon Musk compared the exemplary to Opus, Anthropic LLM designed for intensive and analyzable tasks.

“Based connected beardown affirmative feedback from customers successful our beta trial program, @SpaceXAI volition marque Grok 4.5 disposable to the nationalist tomorrow. It is an Opus-class model, but faster, much token-efficient and little cost,” wrote Musk successful a station connected X.

Musk aboriginal added: “Our interior appraisal is that Grok 4.5 is astir comparable to Opus 4.7, but overmuch faster. The operation of capability, faster velocity and little outgo is what makes it competitive.”

SpaceXAI says that its caller exemplary costs $2 per cardinal input tokens and $6 per cardinal output tokens. That’s rather competitive, if Grok’s capabilities lucifer SpaceXAI’s rhetoric.

Opus 4.7, by comparison, costs $5 per cardinal input tokens, and $25 per cardinal output tokens. OpenAI has tiered costs for antithetic exemplary versions: Sol, its astir expensive, costs $5 for input tokens and $30 for output, portion its slightest expensive, Luna, costs $1 for input and $6 for output.

It’s a large week for AI exemplary releases. OpenAI is planning to release GPT 5.6, its latest, astir almighty model, connected Thursday. The merchandise of that exemplary had antecedently been constricted by the Trump administration, owed to concerns astir its information implications. OpenAI has called it its “strongest exemplary yet.”