Update July 8, 2026: SpaceXAI officially announced Grok 4.5 just hours after Elon Musk\u2019s \u201cCorrect\u201d confirmation. The model is live now in Grok Build, Cursor, and the SpaceXAI API \u2014 not \u201ctomorrow\u201d as originally indicated. Full specs, benchmarks, and pricing details below.
Elon Musk just confirmed Grok 4.5, the biggest AI release of the month, with a single word.
He replied \u201cCorrect\u201d to a post from AI commentator tetsuo detailing everything known about Grok 4.5 going public. The post laid out the full picture. Grok 4.5 is built on xAI\u2019s V9 foundation, a 1.5 trillion parameter model roughly three times the size of the v8-small architecture behind Grok 4.3. It had been in private beta at SpaceX and Tesla since June 28.
Then, hours later, SpaceXAI published the official announcement and dropped the model early. It is live now in Grok Build, Cursor on all plans, and the SpaceXAI console.
What Elon confirmed
The exchange could not be simpler. tetsuo posted a detailed breakdown of Grok 4.5\u2019s specs, training, and release timeline. The thread ended with \u201cTomorrow we get to use it!\u201d Elon replied \u201cCorrect.\u201d
That single word is as close to an official launch announcement as SpaceXAI typically makes. The company does not do press releases the way OpenAI or Google do. Musk\u2019s X posts are the primary communication channel.
The post he confirmed covers the key specs:
Grok 4.5 runs on V9, xAI\u2019s new 1.5 trillion parameter foundation, roughly three times the size of the v8-small architecture behind Grok 4.3. It is the largest model SpaceXAI has shipped. It trained on Blackwell GPUs at the Colossus cluster in Memphis.
Coding is the headline feature. Cursor\u2019s developer data was added in supplemental training, and reinforcement learning has been improving the model daily alongside the Grok Build harness. The next foundation model already in training will have Cursor data included from the start of pre-training.
Musk has previously called it \u201cOpus-level\u201d and said it is faster, more token-efficient, and cheaper than competing models.
The official announcement: Grok 4.5 is SpaceXAI\u2019s first coding-first model
SpaceXAI\u2019s official blog post filled in the details Musk\u2019s one-word confirmation left out. Grok 4.5 is the company\u2019s first model trained specifically for coding and agents, and it was trained alongside Cursor. The model was trained across tens of thousands of NVIDIA GB300 GPUs with heavy investment in data filtering, deduplication, and domain-focused selection.
Benchmarks
The numbers tell a competitive story. Grok 4.5 scores 62% on DeepSWE 1.0 (within SpaceXAI\u2019s own harness), 83.3% on Terminal Bench 2.1, and 64.7% on SWE Bench Pro. It trails Fable (max) and GPT 5.5 (xhigh) on most coding benchmarks but beats Opus 4.8 (max) on Terminal Bench and comes close on SWE Bench Pro. On Harvey\u2019s Legal Agent Benchmark, it scored #1.
The real differentiator is efficiency. Grok 4.5 averages 15,954 output tokens per SWE Bench Pro task, roughly 4.2x fewer tokens than Opus 4.8 (max) at 67,020. That translates directly to cost savings at scale.
Speed and pricing
Grok 4.5 serves at 80 tokens per second. SpaceXAI claims it is faster than flash models from competitors. The pricing is aggressive: $2 per million input tokens, $6 per million output tokens. Combined with the 2x token efficiency advantage, SpaceXAI is positioning Grok 4.5 as the cheapest frontier option per unit of intelligence delivered.
Office work and agentic tasks
Beyond coding, SpaceXAI highlighted Grok 4.5\u2019s capability at office work. The model is the default in Grok Build and can build complex Excel models with web research, multi-sheet formulas, and even sticky notes. In PowerPoint and Word, it uses native shapes for diagrams and writes clean prose. SpaceXAI announced plugins for Word, PowerPoint, and Excel.
First flagship under SpaceXAI
This is the first flagship model release since xAI rebranded to SpaceXAI earlier this week. The merger finalized on July 6, and Grok 4.5 is the first major product to ship under the new brand.
The timing is aggressive. xAI\u2019s roadmap calls for a from-scratch foundation model shipping every month through the end of 2026. Grok 4.3 launched in April. Grok 4.5 follows in July. If they hold that cadence, expect Grok 5 variants before the year is out, with some projections reaching 10 trillion parameters.
SpaceXAI has also been expanding its product line beyond the model itself. The Grok Voice Agent Builder launched in beta last week, a no-code voice platform at $0.05 per minute with telephony tools and guardrails. A SpaceX AI device prototype was shown to investors. The model release is the engine; the ecosystem around it is growing fast.
What Grok 4.5 changes today
Grok 4.5 is live now. Anyone can test the \u201cOpus-level\u201d claims today in Grok Build (free for a limited time) or in Cursor on any plan. API access is available from the SpaceXAI console at $2/$6 per million tokens.
The competitive landscape just shifted in real time. A 1.5 trillion parameter model focused on coding and agents, trained on GB300 GPUs with Cursor data, served at 80 TPS, and priced aggressively enough to undercut most of the market is not a minor release. It is SpaceXAI\u2019s strongest statement yet that it intends to compete on capability, speed, and price simultaneously.
EU users will have to wait. Grok 4.5 is not yet available in the EU in any SpaceXAI products or the API console. EU availability is expected in mid-July.
Bottom line
Grok 4.5 dropped early, and it is live now. SpaceXAI\u2019s first coding-first model brings competitive benchmarks, industry-leading token efficiency, and aggressive pricing in a single package. The first flagship under the new brand shows SpaceXAI is playing to win on developer workloads.
I will be testing it against real workflows and sharing results. The \u201cCorrect\u201d claim has benchmarks behind it now. The question is whether it holds up in the wild.
