Musk’s SpaceXAI Launches Grok 4.5, Calling It an ‘Opus-Class’ Rival to Anthropic

SpaceXAI unveiled Grok 4.5 on Wednesday, calling it its smartest model yet and its strongest for coding, agentic tasks, and general knowledge work.

The release marks the company’s first model since going public on the Nasdaq under ticker SPCX in June, and the first built with Cursor, the AI coding tool SpaceX acquired for $60 billion in an all-stock deal weeks earlier.

On X, Musk compared Grok 4.5 directly to Anthropic’s flagship models, saying it’s “an Opus-class model, but faster, more token-efficient and lower cost,” before adding that internal testing places it “roughly comparable to Opus 4.7, but much faster.”

How Grok 4.5 Stacks Up on Paper

SpaceXAI’s benchmark data shows Grok 4.5 performing competitively, though not always leading. 

On the DeepSWE 1.0 coding benchmark, it scored 62%, behind Anthropic’s Fable (66.1%) and OpenAI’s GPT-5.5 (64.31%), but ahead of Opus 4.8 (55.75%). 

On the tougher SWE Bench Pro benchmark, Grok 4.5 reached 64.7%, trailing Fable’s 80.4% and Opus 4.8’s 69.2%. 

SpaceXAI says Grok 4.5’s advantage is efficiency, claiming 80 tokens per second while solving the average SWE Bench Pro task with about 15,954 output tokens—roughly 4.2× fewer than Opus 4.8’s 67,020. 

That efficiency supports its lower pricing of $2/$6 versus Opus 4.7’s $5/$25 and OpenAI’s top-tier $5/$30, a cost advantage SpaceXAI is betting will appeal to developers as token costs continue rising.

Built With Cursor, Aimed at the Whole Office

Beyond coding, SpaceXAI says Grok 4.5 can build complex Excel models with web research and multi-sheet formulas, generate PowerPoint presentations with native slide shapes, and draft Word documents through dedicated Microsoft plugins. 

The company also said the model scored highest on Harvey’s Legal Agent Benchmark, a test geared toward evaluating AI for legal tasks and industry-specific workflows. 

Grok 4.5 is now the default model in Grok Build, SpaceXAI’s own coding agent, and is available immediately inside Cursor on all subscription plans, alongside SpaceXAI’s developer console for API access. 

Reuters also noted that Cursor helped train the model following SpaceX’s acquisition of its parent company, Anysphere. SpaceXAI said the model isn’t yet available in the European Union, with access expected to arrive there in mid-July.

A Crowded Week for Frontier AI

Grok 4.5’s launch comes just ahead of OpenAI’s expected GPT-5.6 release on Thursday, a model the company calls its strongest yet after the release had previously been delayed amid Trump administration concerns over its security implications. 

With SpaceXAI, Anthropic, and OpenAI all releasing major updates within days of each other, Grok 4.5 further intensifies the AI race, where cost and speed are increasingly marketed alongside raw benchmark performance. 

Source: Introducing Grok 4.5