SpaceXAI released Grok 4.5 on Wednesday, its first model since xAI formally dissolved into SpaceX earlier this year and the company’s first release built jointly with Cursor, the coding assistant startup SpaceX agreed to acquire for $60 billion in a deal expected to close in the third quarter.
Elon Musk described it on X as an “Opus-class model,” a direct reference to Anthropic’s flagship line, adding that it’s faster, more token-efficient, and cheaper to run, while privately assessing it as roughly comparable to Opus 4.7 rather than Anthropic’s newest release.
How it stacks up on performance and price
SpaceXAI’s own benchmark disclosures put Grok 4.5 at 62% on DeepSWE 1.0, 83.3% on Terminal Bench 2.1, and 64.7% on SWE Bench Pro, trailing Anthropic’s Fable (max) and OpenAI’s GPT 5.5 on most of those same tests while running ahead of Anthropic’s Opus 4.7.
Where it does lead is SWE Marathon, a longer-horizon resolution benchmark, at 29%. Grok 4.5 also stands out on cost and speed.
It processes up to 80 tokens per second and uses about 4.2 times fewer output tokens than Opus 4.8 on comparable tasks. SpaceXAI prices the model at $2 per million input tokens and $6 per million output tokens, compared with Opus 4.8’s $5 and $25.
Independent evaluator Artificial Analysis ranked it fourth on its GDPval-AA v2 real-world task leaderboard, behind Anthropic’s most recent releases, at roughly $0.49 per completed task.
Why this release matters beyond the numbers
The Cursor tie-in is the more strategically interesting piece. Training a flagship model alongside one of the most widely used coding tools on the market gives SpaceXAI direct distribution into developer workflows rather than competing purely as a standalone chatbot.
There’s also a quieter tension worth watching: SpaceXAI has been leasing its GB300 compute capacity to Anthropic and Google, and as its own model training needs grow, it may increasingly have to choose between running its own workloads and continuing to rent that capacity to the same companies Grok is now positioned against.
The timing adds pressure too. Grok 4.5 landed a day before OpenAI’s scheduled public launch of GPT-5.6, making this one of the more crowded model-release weeks of the year.
Grok 4.5 is live today in Grok Build, across all Cursor plans, and through the SpaceXAI API console, though EU users will have to wait until mid-July for access.
Musk has framed the release as an opening move rather than a finishing one, saying SpaceXAI expects to close the remaining performance gap with its larger competitors soon.
Whether that happens matters less right now than whether the price and speed advantage is enough to pull developer workflows away from entrenched tools while SpaceXAI works on catching up on raw capability.
Source: Official xAI announcement, "Grok 4.5"
