SpaceXAI Launches Grok 4.5: Inside the 1.5-Trillion-Parameter V9 Model Trained on Cursor Data

Here is a fact-checked summary of what SpaceXAI announced regarding Grok 4.5 on July 9, 2026.

What SpaceXAI Announced

On July 8, 2026 (the launch announcement was Wednesday July 8, with public availability rolling out on July 9), SpaceXAI launched Grok 4.5, calling it the company's "strongest model ever" and its "smartest" AI model to date . The launch was timed to coincide with OpenAI rolling out a new ChatGPT model on the same day .

Release Timeline

• June 28, 2026 — Elon Musk announced Grok 4.5 entered private beta at Tesla and SpaceX, based on the 1.5-trillion-parameter V9 foundation model with Cursor data added in supplemental training .
• June 30, 2026 — Musk confirmed Grok 4.5 would exit private beta and become publicly available "tomorrow" .
• July 8, 2026 — SpaceXAI officially launched Grok 4.5 to the public, making it the default model in Grok Build and SpaceXAI's coding and engineering platform .
• July 9, 2026 — Full public availability across subscription tiers and the API .

Architecture (V9 Foundation, ~1.5 Trillion Parameters)

Elon Musk confirmed on X that Grok 4.5 is "based on our 1.5T V9 foundation model" — roughly three times the scale of the V8-small architecture that underpinned earlier Grok 4 variants . Key details:

• 1.5 trillion parameters on the V9 foundation model, the first use of the V9 architecture .
• Trained across tens of thousands of Nvidia GB300 GPUs with a focus on meticulous data filtering, deduplication, and quality scoring .
• Early internal benchmarks at Tesla and SpaceX showed performance "close to, perhaps exceeding Opus" (Anthropic's Claude Opus class) .
• Positioning as an "Opus-class model" — faster, better token efficiency, and lower cost than competing top-tier models .

Performance Positioning

Musk explicitly described Grok 4.5 as an "Opus-class model" — directly positioning it against Anthropic's top-tier Claude Opus while emphasizing it is faster and more cost-efficient .

Note on claimed rivals: The specific model names "GPT 5.5" and "Opus 4.8" that you referenced in the prompt did not appear in the sources gathered. Sources instead describe Grok 4.5 as competing against Anthropic's Claude Opus (current top Opus tier) and OpenAI's newest GPT model (which launched the same day, July 8) . The search budget was exhausted before these specific model version numbers could be independently verified. The broad competitive context is confirmed — SpaceXAI positioned Grok 4.5 directly against OpenAI's and Anthropic's flagship models — but the precise version strings (GPT-5.5, Opus 4.8) are not corroborated by the available source evidence.

Cursor Data and the $60 Billion Acquisition

SpaceX formally agreed to acquire Anysphere, the parent company of the AI coding agent Cursor, for $60 billion in an all-stock transaction, announced on June 16, 2026 . The deal was the largest acquisition of a venture-backed startup ever .

Grok 4.5 is the first model built with training data from Cursor. How Cursor data was used:

• Supplemental training — Cursor's developer session data was added as supplementary training data on top of the V9 base model, not training the V9 foundation from scratch .
• Coding sharpening — Real developer interaction data (code completions, edits, debugging sessions) baked into supplemental training to sharpen the model's coding and agentic task capabilities .
• Both companies explicitly partnered: Cursor said "We've partnered with SpaceXAI to train Grok 4.5" .
• Cursor also released Grok 4.5 within its own platform on July 8, calling it "our most intelligent model and the first we've built for more than software engineering" .
• SpaceX deployed "a few dozen" top Starlink and Starship engineers plus Cursor engineers to work on the new foundation model .

Availability and Pricing

Consumer subscription access (confirmed by multiple sources) :

• X Premium+ — $40/month (or $395/year on web) — includes full Grok 4.5 access bundled with X platform features .
• SuperGrok — $30/month (or $300/year) — standalone Grok subscription, the cheapest standalone path to full Grok access .
• SuperGrok Lite — $10/month — entry-level paid tier launched March 25, 2026 .
• SuperGrok Heavy — $300/month — premium tier for earlier top-tier Grok access (historically for Grok 4 Heavy) .

xAI Developer API pricing (from official docs) :

Model	Context Window	Input Price (per 1M tokens)	Cached Input	Output Price (per 1M tokens)
grok-4.5	500k tokens	$2.00	$0.50	$6.00
grok-4.3 (predecessor)	1M tokens	$1.25	—	$2.50

Predecessor pricing for reference:

• Grok 3 / Grok 4 — previously available via the same subscription tiers (SuperGrok $30/mo, X Premium+ $40/mo, SuperGrok Heavy $300/mo for Grok 4 Heavy) .
• Grok 4 Heavy previously required the $300/month SuperGrok Heavy tier .
• The xAI API previously priced grok-4.3 at $1.25 input / $2.50 output per 1M tokens, making Grok 4.5 significantly more expensive on output ($6.00 vs $2.50) .

Competitive Context

Same-day launches:

• OpenAI rolled out a new ChatGPT model on the same day (July 8, 2026), though the specific version "GPT-5.6" could not be independently verified from sources .
• Google / Gemini 3.5 Pro and Anthropic / Fable 5 — These specific model names were not found in the available source set. The search budget was exhausted before dedicated queries could be run for these competitors.

What is confirmed:

• Grok 4.5 was explicitly marketed as a coding and agentic-work tool, more than a consumer chatbot — positioned against Anthropic's Claude Opus and OpenAI's latest flagship .
• Musk claimed Grok 4.5 is "maximally truth-seeking" and shows early performance near or exceeding Opus, while being faster and cheaper to run .
• The model is described as excelling at coding, finance, and completing tasks without human intervention .

Key Caveats

• No independent benchmarks had been published at launch — all performance claims were self-reported by SpaceXAI .
• The specific rival model version strings (GPT-5.5, Opus 4.8, Gemini 3.5 Pro, Fable 5) in the prompt are not substantiated by the sources collected. The available evidence confirms the broad competitive positioning but not those exact model names.
• The Cursor acquisition closed at a $60 billion valuation in an all-stock deal, but the transaction may still be pending final regulatory completion .