Grok 4.5, Musk's 'Opus-class' model starts at $2 per million tokens

In recent hours, SpaceXAI introduced Grok 4.5, the most capable model in its catalog and the first released after the IPO and acquisition of Cursor, the programming startup valued at $60 billion in the deal.

Based on strong positive feedback from customers in our beta test program, @SpaceXAI will make Grok 4.5 available to the public tomorrow. It is an Opus-class model, but faster, more token-efficient, and lower cost. — Elon Musk (@elonmusk) July 8, 2026

"It is an Opus-class model, but faster, more token-efficient, and less expensive," Musk wrote on X, referencing the Opus family of Anthropic, the flagship models for complex tasks. Grok 4.5 is aimed at programming and agency tasks, not for casual chatbot use.

Price Leverage

SpaceXAI has set Grok 4.5 at $2 per million input tokens and $6 for output. For comparison, Anthropic's flagship Opus model costs $5 and $25, while OpenAI's offerings span multiple tiers: GPT-5.6 Luna, the cheapest, is priced at $1 and $6, while Sol is at $5 and $30. Grok positions itself below the pricier rivals at a time when companies are closely monitoring token spending.

The company also claims a token efficiency double that of other benchmark models. If this holds up in real usage, it would be a tangible advantage, but it remains a manufacturer’s statement and is currently not an independently verified value.

A chart released with the announcement supports that Grok 4.5 surpasses the flagship Opus model on several benchmarks. However, Musk himself downplays this: the internal valuation assesses it as "roughly comparable to Opus 4.7, but much faster." The company admits to outpacing some models from OpenAI and Anthropic on speed, price, and performance, but not their largest and latest, and hopes to bridge the gap soon.

A Model for Programmers and Wall Street

In addition to code, Grok 4.5 also targets legal and financial work, adding cybersecurity functionalities. The push toward finance is deliberate: Musk had admitted this year that his AI unit, known as xAI before merging with SpaceX, was lagging behind in programming. Since then, the team has been rebuilt, and the company has courted Wall Street clients for the Grok chatbot. Grok 4.5 is the clearest signal of this pivot towards paying enterprise customers.

The model is already available in Grok Build, within Cursor on all plans, and through the SpaceXAI console; broader public release is expected today. However, the model is not yet accessible in the European Union.

Regulators' attention weighs on the entire industry, focusing on the cybersecurity risks of new models; Cursor has stated that it has adopted measures to "detect and block malicious actors" while preserving legitimate security research.

SpaceXAI has trained Grok 4.5 on the same computing power it rents to rivals Anthropic and Google. As its models become more demanding, the company will need to decide whether to feed them or resell that capacity for liquidity. For now, Musk bets that a cheaper Grok designed for programmers can capture clients, while still trailing behind the best models in pure power.