SpaceXAI’s New Grok 4.5 Slashes AI Coding Costs by 75 Percent

SpaceXAI (formerly xAI until it was rebranded by SpaceX) has rolled out Grok 4.5 to the public, a new AI model built specifically to handle serious coding and engineering work rather than casual chatbot conversations.

Grok 4.5 runs on the company’s new V9 foundation, which packs 1.5 trillion parameters. What does that mean? That’s a measure of how big and capable the model is, and it makes Grok 4.5 about three times larger than the v8-small setup behind Grok 4.3. To train something this size, SpaceXAI leaned on its Colossus supercluster in Tennessee running NVIDIA’s current flagship Blackwell chips. The payoff is a model that’s efficient enough to keep the cost of running heavy tasks fairly low.

Pricing is what makes this new model super competitive. Grok 4.5 costs $2 per million input tokens and $6 per million output tokens, and that noticeably undercuts Anthropic’s flagship, Claude. For comparison, Claude Opus 4.8 runs $5 per million input tokens and $25 per million output tokens. So on the output side, where the bulk of the cost usually lands, Grok 4.5 comes in at roughly a quarter of the price. OpenAI’s lineup sits in between, with its priciest model at $5 input and $30 output per million tokens, while its cheaper option nearly matches Grok at $1 input and $6 output.

On the benchmarks, Grok 4.5 scored 64.7 per cent on SWE-Bench Pro, a tough test for real software engineering work. And on Cursor’s (owned by SpaceX now) CursorBench platform, it’s handling automated coding tasks for just $1.51 each.

So what’s the big deal? For businesses, Grok 4.5 signals a move toward cheaper AI agents that can work at scale, and it puts pressure on rivals to explain why they charge so much more. The strong performance at a low price means companies can run automated engineering work for a fraction of what it used to cost.

Elon Musk pitched it as a model that delivers top-tier performance while running faster and cheaper than the competition. He also noted that Grok 4.5 isn’t yet using special software the company is still building to help it run more efficiently. Once that software is ready and fine-tuned to squeeze the most out of NVIDIA’s latest AI chips (the GB300, its most powerful setup to date), Musk says the model could run twice as fast or better.

The model is ready to use right away across a bunch of developer tools. Engineers can tap into it through Cursor, the Grok Build command line, Vercel, and the official developer API. People on the OpenClaw platform can also link their existing premium subscriptions to access it without a separate update.

For consumers, cheaper AI that codes well tends to trickle down, meaning the apps and services you use could get built faster, updated more often, and cost less to run. A price war at the top also tends to drag down what regular users pay for chatbots and coding assistants, since when one company cuts rates, the others usually follow. Competition is a great thing for consumers and this AI arms war is just beginning it seems.