SpaceX AI Releases Grok 4.5, Pricing Significantly Lower Than Anthropic
Alina Collins
Musk's SpaceXAI released Grok 4.5 on Wednesday, pricing output tokens at roughly one-quarter of Anthropic's Claude Opus 4.8 — a direct price-war play for the enterprise AI coding market.
How much cheaper — and compared to whom?
Grok 4.5 output costs $6 per million tokens. Anthropic's Claude Opus 4.8 charges $25 — a gap of roughly 4×.
On the input side the spread is narrower: Grok 4.5 at $2/M vs. Opus 4.8 at $5/M.
OpenAI's GPT-5.6 Luna matches Grok 4.5 on output ($6) but undercuts it on input ($1). This means → Grok 4.5 is not positioning as the cheapest model overall, but as "Opus-level capability at a mid-low price."
What is Musk actually claiming?
On X, Musk called Grok 4.5 "an Opus-class model, but faster, more token-efficient, and cheaper."
In plain terms = he wants the market to believe you can get near-Anthropic-flagship performance for one-quarter the price.
That pushes "value for money" — not absolute benchmark dominance — to the center of the launch narrative.
What supports the training and ecosystem?
Grok 4.5 was trained on tens of thousands of Nvidia GB300 GPUs, with SpaceXAI emphasizing refined data filtering, deduplication, and quality scoring.
AI coding tool Cursor confirmed it co-trained Grok 4.5. SpaceX last month agreed to acquire Cursor's parent Anysphere in an all-stock deal worth roughly $60 billion.
This reflects an effort to lock model training and downstream coding tools into a single "model + tool" closed-loop ecosystem.
Where can you use it now?
Grok 4.5 is live through three channels: SpaceXAI's own coding tool Grok Build, Cursor, and the SpaceXAI developer console (API key required).
EU access is expected by mid-July.
OpenAI plans to publicly release GPT-5.6 on Thursday — a launch previously delayed over U.S. government national-security concerns. This means → Grok 4.5 went live one day ahead, competing for an extremely narrow attention window.
Can the low-price playbook work?
Musk is making a clear bet: when enterprises choose coding and agentic tools, price sensitivity outweighs benchmark rankings.
But Grok 4.5 has no published third-party benchmark results so far. "Opus-class" remains SpaceXAI's own claim, not an independent verdict.
Put simply = a low price solves the "will they try it" problem. "Will they stay" depends on real-world performance — and that is still unproven.
Content is for reference only, not financial advice.