xAI launches Grok 4.5: Musk's model promises Opus-level performance at a fraction of the cost

July 9, 2026 Daniel Cesak

SpaceXAI — the new name of the company that until recently operated as xAI — officially unveiled the Grok 4.5 model yesterday. Elon Musk called it an "Opus-class model" and promises performance comparable to the best models from Anthropic, but with double the token efficiency and a significantly lower price. The new version of Grok arrives just a few weeks after SpaceX's historic stock market debut and just days before OpenAI is set to release its most anticipated model, GPT-5.6.

Don't call Grok xAI anymore. It's now SpaceXAI

The rebranding of xAI to SpaceXAI was completed on July 6, 2026 — three days before the release of Grok 4.5. The company officially gave up its standalone identity and merged under the SpaceX umbrella, placing Musk's AI division in an entirely new context. SpaceX is no longer just a rocket company, but a technology conglomerate that also includes the social network X (formerly Twitter) and Anysphere — the developer of the popular AI editor Cursor, whose $60 billion acquisition was announced by SpaceX on June 16.

Cursor is a strategic lever for SpaceXAI. While Anthropic's Claude Code and OpenAI's Codex have dominated agentic programming for months, Musk's own tool Grok Build, launched in May as an "early beta," has yet to impress. By acquiring Cursor, SpaceXAI gains immediate access to millions of developers and technology that ranks among the most widely used AI coding assistants.

Grok 4.5: What the new model can do and for how much

SpaceXAI published a set of benchmarks showing Grok 4.5 sitting just below the top tier — trailing models like Anthropic's Claude Fable 5 or the upcoming GPT-5.6 Sol from OpenAI, but holding a competitive position. The company particularly emphasizes "twice the token efficiency" compared to other leading models, a claim that — if proven in practice — could represent a significant advantage for enterprise customers.

The key argument is price. Grok 4.5 costs $2 per million input tokens and $6 per million output tokens. For comparison:

Claude Opus 4.7 (Anthropic): $5 / $25 per million tokens
GPT-5.6 Sol (OpenAI, top-tier version): $5 / $30 per million tokens
GPT-5.6 Luna (OpenAI, base version): $1 / $6 per million tokens

If Grok 4.5's performance truly matches what SpaceXAI promises, we get a model that is 2.5× cheaper on input and more than 4× cheaper on output than Opus 4.7. For companies with high API call volumes, this would mean dramatic cost savings.

Comparison with the competition: A big week for AI

Grok 4.5 isn't entering an empty market — quite the opposite. This week is one of the most packed in the history of the AI industry. OpenAI on Thursday, July 10 will release the trio of GPT-5.6 models (Sol, Terra, Luna), which were originally held back by the Trump administration over safety concerns. The administration gave the green light only yesterday.

Meanwhile, Anthropic has in recent weeks rolled out Claude Cowork — an agentic tool that expanded from desktop to mobile and web — and Meta launched Muse Spark 1.1 just days ago, a model focused on programming.

Elon Musk on X openly compared Grok 4.5 to Anthropic's Opus model: "Our internal evaluation shows that Grok 4.5 is roughly comparable to Opus 4.7, but significantly faster. The combination of capabilities, higher speed, and lower price is what makes it competitive."

Availability in Czechia and Europe

Grok 4.5 is available through the Grok.com web interface and via the SpaceXAI API. To access the model, you need an X Premium+ or SuperGrok subscription. Czech users have unrestricted access to Grok — unlike the Grok implementation in European Tesla vehicles, which still bypasses the Czech Republic. In February 2026, Tesla made Grok available in nine European countries (United Kingdom, Ireland, Germany, Switzerland, Austria, Italy, France, Portugal, Spain), but Czechia was not among them and still isn't.

The model currently does not support Czech as a primary language — communication takes place primarily in English, although Grok can handle multilingual queries. For Czech companies and developers, however, the key factor is primarily the low API price, which could make it an attractive choice for automation and enterprise deployment.

What this means for the average user

The practical impact of Grok 4.5's release is twofold. First, the price war among AI giants is intensifying. When a model with performance close to Opus costs a third to a quarter of the price, it pressures other players to lower their prices. For end users, this means cheaper and more accessible AI tools.

Second, the integration with Cursor — once the acquisition is completed in the third quarter of 2026 — could make Grok the default model for millions of developers. This would fundamentally change the dynamics of the AI-assisted programming market, where Claude (Anthropic) and ChatGPT/Codex (OpenAI) currently dominate.

How is Grok 4.5 different from the previous Grok 3 version?

Grok 4.5 primarily brings higher benchmark performance, better token efficiency, and lower operating costs. Compared to Grok 3, it has improved in coding, logical reasoning, and the ability to handle complex tasks. It is the first model released under the SpaceXAI banner after the rebranding was completed.

Can I use Grok 4.5 for free?

A free version of Grok exists, but it has limited features. Full access to Grok 4.5 requires an X Premium+ subscription (approximately 400 CZK per month) or the SuperGrok plan. API prices are then charged based on token consumption — $2 per million input and $6 per million output tokens.

When will Grok 4.5 be integrated into Cursor?

SpaceX announced that the acquisition of Anysphere (Cursor) will be completed in the third quarter of 2026. A specific date for Grok 4.5's integration into Cursor has not yet been announced, but it is expected that SpaceXAI will want to incorporate its model into the popular editor as soon as possible after the acquisition is finalized.