SpaceXAI released Grok 4.5 to the public on July 8, 2026, eleven days after Elon Musk first announced it in private beta at SpaceX and Tesla. It's live now in Grok Build, inside Cursor on all plans, and through the SpaceXAI console. Musk is calling it "an Opus-class model, but faster, more token-efficient and lower cost." xAI's own published benchmark chart tells a more mixed story than that framing suggests.
Grok 4.5 is xAI's new flagship model, built on the 1.5-trillion-parameter V9 foundation and trained alongside Cursor, the AI coding editor SpaceX agreed to acquire for $60 billion in June 2026. It's priced at $2 per million input tokens and $6 per million output tokens, runs at roughly 80 tokens per second, and is now the default model inside Grok Build. It is not yet available in the EU.
What Actually Shipped
This is the first model to reach any kind of public release from xAI's V9 architecture, the successor to the V8-small foundation that powered earlier Grok 4 builds. V9 finished its primary training run on May 26, 2026, entered private beta at SpaceX and Tesla on June 28, and moved to public availability today, a turnaround of just over a week from internal-only to open access.
Three things distinguish this release from a routine point-update:
Cursor-trained. xAI folded real developer session data from Cursor into supplemental training: debugging traces, multi-file diffs, and user corrections, not just static code corpora. That's a different signal than most coding models train on.
Token efficiency is the actual headline, not raw score. On SWE-Bench Pro, xAI reports Grok 4.5 resolves tasks using an average of 15,954 output tokens, against 67,020 for Opus 4.8 (max) on the same benchmark — a 4.2x gap. If that number holds up under independent testing, it's a bigger deal for anyone paying by the token than a benchmark win.
It's the first release since SpaceX's Cursor acquisition and February 2026 xAI rollup, which is why the marketing leans hard into "coding and agentic work" rather than consumer chatbot framing.
The Benchmark Chart, Read Straight
xAI published four benchmark comparisons on its own launch page. Here they are exactly as posted, not summarized:
Benchmark | Fable (max) | GPT-5.5 (xhigh) | Opus 4.8 (max) | Grok 4.5 | GLM-5.2 |
|---|---|---|---|---|---|
DeepSWE 1.0 | 66.1% | 64.31% | 55.75% | 62.0% | — |
DeepSWE 1.1 | 70% | 67% | 59% | 53% | 44% |
Terminal-Bench 2.1 | 84.3% | 83.4% | 78.9% | 83.3% | — |
SWE-Bench Pro | 80.4% | — | 69.2% | 64.7% | 62.1% |
Two things worth being direct about, since most coverage today is just repeating the "Opus-class" quote without opening the chart:
Grok 4.5 beats Opus 4.8 on two of the four benchmarks xAI chose to publish (DeepSWE 1.0 and Terminal-Bench 2.1), and loses to it on the other two (DeepSWE 1.1 by 6 points, SWE-Bench Pro by 4.5 points). "Opus-class" is a defensible label for a model that splits four head-to-heads roughly evenly. It's not the same claim as "beats Opus," which is closer to how the quote is being repeated across the launch-day news cycle.
Claude Fable 5 leads on all four benchmarks shown, including against Grok 4.5, GPT-5.5, and Opus 4.8. xAI included that comparison in its own chart, which means the launch material is transparent about where the model actually stands in the field rather than only picking favorable comparisons. Worth knowing if a benchmark chart is what's driving a purchase decision here.
None of these four benchmarks have independent, third-party-verified numbers yet for Grok 4.5. They're xAI's self-reported figures from its own launch page. That's normal for a same-day release, but it means today's number should be read as "the vendor's number," not a settled result.
Pricing and Access
Grok 4.5 | |
|---|---|
Input tokens | $2 / million |
Output tokens | $6 / million |
Speed | ~80 TPS |
Access today | Grok Build (default model), Cursor (all plans), SpaceXAI console |
EU availability | Not yet |
For context, that pricing sits below where Opus-tier models have typically been positioned, which is consistent with xAI's repeated public pattern of undercutting Anthropic and OpenAI on cost while trailing on raw benchmark leadership.
Where This Actually Helps
Grok 4.5 is being positioned specifically for coding, agentic tasks, and office work rather than general chat. The concrete use cases xAI is highlighting:
End-to-end app builds from a single prompt — xAI's own examples show a full Three.js solar-system simulation with adjustable time controls and a styled HUD, generated from one instruction.
Excel work that involves web research, multi-sheet formulas, and left-behind notes for future reference.
Word and PowerPoint, including native PowerPoint shapes for diagrams and slide layout, not just text generation.
Harvey's Legal Agent Benchmark — Grok 4.5 scores #1 here, which is a notable claim for a model marketed primarily around coding, and suggests the training mix reaches further than engineering tasks.
If your work is inside Cursor already, this is a same-day, zero-friction option to try since it's live on all plans there now. If you're evaluating on raw coding benchmark ceiling rather than cost, the chart above is the one to look at before switching anything.
FAQs
Is Grok 4.5 publicly available? Yes, as of July 8, 2026. It shipped in Grok Build (as the default model), inside Cursor on all plans, and through the SpaceXAI console. It is not yet available in the EU.
Does Grok 4.5 actually beat Claude Opus 4.8? Mixed, by xAI's own published numbers. It beats Opus 4.8 on DeepSWE 1.0 and Terminal-Bench 2.1, and loses to Opus 4.8 on DeepSWE 1.1 and SWE-Bench Pro. "Opus-class" is accurate as a tier description; "beats Opus" isn't fully supported by the same chart xAI released.
What is Grok 4.5 built on? xAI's V9 foundation model, 1.5 trillion parameters, roughly 3x the scale of the V8-small architecture behind earlier Grok 4 releases. Training was supplemented with real developer session data from Cursor.
How much does Grok 4.5 cost? $2 per million input tokens, $6 per million output tokens, served at roughly 80 tokens per second.
What is SpaceXAI? The renamed entity following SpaceX's acquisition of xAI in February 2026. SpaceX also agreed to acquire Cursor for $60 billion in June 2026, which is why Grok 4.5's training and launch are so tightly tied to Cursor.
Sources
Benchmark figures, pricing, and feature details verified against xAI's official Grok 4.5 launch page.
