SpaceXAI Launches Grok 4.5 With Pricing That Reshapes the AI Market

SpaceXAI Launches Grok 4.5 With Pricing That Reshapes the AI Market

SpaceXAI released Grok 4.5 on Wednesday, and the model is drawing attention for two reasons. It posts competitive scores on benchmarks for coding, legal work, and office productivity. More pointedly, its pricing undercuts every major rival by a wide margin.

The company was known as xAI until a recent rebranding. It trained Grok 4.5 on tens of thousands of Nvidia GB300 GPUs in partnership with Cursor, a coding startup it is acquiring for roughly $60 billion. SpaceXAI says it focused on data quality over raw volume, investing heavily in filtering, deduplication, and domain-specific curation. The model was further refined through reinforcement learning, the same technique used by OpenAI and DeepSeek to teach models chain-of-thought reasoning.

Pricing That Gets Procurement's Attention

Grok 4.5 is priced in the range of $2 per million tokens for both input and output. By comparison, Anthropic's Claude Opus runs closer to $15 per million tokens, and OpenAI's GPT-5.6 models sit in the $10 to $15 range. The difference is large enough to change budget conversations inside most organisations.

For a company running thousands of AI queries each day, that gap compounds fast. A workflow that costs $100 on Claude Opus might cost less than $15 on Grok 4.5. Over a quarter, the savings can cover headcount or infrastructure costs elsewhere.

Musk has said the model competes with "last year's Claude Opus." That framing acknowledges a gap in raw capability but argues the price-performance ratio more than compensates. Whether the trade-off works depends on the job. For sensitive legal analysis or high-stakes financial modelling, accuracy matters more than cost. For code generation, document drafting, and data extraction, the lower price is hard to argue with.

Can Grok Shed Its Baggage?

Grok carries a history that makes some enterprise buyers cautious. Earlier versions of the model were used to generate non-consensual deepfake images. At one point Grok adopted an alarming persona called "MechaHitler" that drew widespread criticism and led to bans in some regions.

SpaceXAI says Grok 4.5 has addressed those issues through updated safety training and content filtering. The company points to the model's performance on Harvey's Legal Agent Benchmark, where it claims the top spot. Harvey is a legal AI platform used by major law firms, and its benchmarks test document analysis, contract review, and legal reasoning.

The model can also handle office productivity tasks through a feature called Grok Build. SpaceXAI says it can construct complex Excel models that pull data from the web, work across multiple sheets, and leave notes for the next person reviewing the file. For knowledge workers, that kind of capability could automate hours of repetitive spreadsheet work.

The Cursor Deal Changes Distribution

The $60 billion Cursor acquisition is more than a talent grab. Cursor is one of the most popular AI coding environments among developers. By embedding Grok 4.5 directly into Cursor, SpaceXAI gets instant distribution to a large and active user base.

Developers already using Cursor for code generation can access Grok 4.5 without changing their workflow. That matters. Many engineers are reluctant to adopt new tools, but an upgrade within a familiar interface faces less resistance. It also gives SpaceXAI a real-time feedback loop from the kinds of users who stress-test models hardest.

AI Model Prices Are Falling Fast

Grok 4.5 is part of a broader trend. Inference costs across the industry have been dropping for months. Google's Gemini models, DeepSeek's open-weight releases, and Meta's Llama family all compete on price. OpenAI introduced GPT-5.6 with multiple tiers last week, and Anthropic has been reducing Claude pricing as well.

What makes Grok 4.5 different is the speed of the drop. A model with competitive benchmark scores at this price point would have been unthinkable six months ago. The market is moving from a world where capability was the only consideration to one where price is becoming a decisive factor.

For Australian businesses evaluating AI adoption, this is good news. Lower inference costs make it easier to justify experimentation and production deployment. We recommend running small proofs of value against at least two providers before committing to any single platform. The pricing environment is shifting quickly, and the cheapest option today may not be the cheapest next quarter.

Most organisations now face a procurement question rather than an adoption question. The old model of picking one vendor and standardising across the company may no longer make sense. With prices changing weekly and new capabilities appearing monthly, a flexible multi-model strategy looks smarter than a single-platform bet. We are advising clients to build their AI procurement around an abstraction layer that lets them switch models as market conditions change. That way they capture the price drops without rebuilding their applications.

Grok 4.5 is available now through the Cursor editor for coding tasks and through the SpaceXAI web interface for general use. The company has published documentation and a system card with safety details. Whether the model lives up to its benchmarks outside of controlled testing is something the market will determine over the coming weeks.