xAI Launches Grok 4.5, a Cheaper Rival to Claude Opus for Coding Tasks

xAI, Elon Musk's company, released the new Grok 4.5 model on July 8, marketed as an equivalent to Anthropic's Claude Opus 4.7, but significantly cheaper and faster. Full access rolled out a day later, on July 9, though not yet for users in the European Union.

Opus for a Fraction of the Price

Musk is positioning Grok 4.5 directly against Anthropic's most expensive model. Opus 4.7 is priced at $5 per million input tokens and as much as $25 per million output tokens, while Grok 4.5 costs $2 and $6, respectively. That puts the output-price gap at more than fourfold.

Our internal assessment is that Grok 4.5 is roughly comparable to Opus 4.7, but much faster - Elon Musk, xAI

The company also touts twice the token efficiency compared to competing models, which in practice means lower bills for the same tasks at similar answer quality. The model is meant to be judged primarily on coding, app-building, and office-work performance rather than consumer conversation.

Trained on GB300s, Built with Cursor

Grok 4.5 was trained on tens of thousands of Nvidia GB300 chips, the company's newest generation, using stabilization techniques tailored to very large training runs. xAI also said the model was developed in partnership with Cursor, the popular AI-assisted code editor, and is available across all of that tool's plans starting on launch day.

It's another sign that AI labs are increasingly designing their flagship models with specific coding tools in mind rather than general-purpose chat. Availability in Grok Build, the xAI console, and Cursor from day one is meant to draw in developers accustomed to Claude Code or Codex.

No Access in Europe Yet

Grok 4.5 has not yet reached users in the European Union; xAI says it won't be available there until mid-July. The company hasn't explicitly explained the delay, but it coincides with the August 2 deadline, when the European Commission begins enforcing strict obligations for developers of the most powerful AI models under the EU's AI Act.

For Polish companies and developers, that means another model whose full availability will take longer to arrive than it does for users in the US. Delayed EU rollouts have become the norm in recent months among major AI labs, which test compliance with local regulatory requirements before launching.

The Price War Continues

The Grok 4.5 launch fits into a broader trend of leading labs aggressively cutting token prices. A day earlier, OpenAI released the GPT-5.6 family in three pricing tiers, from $1 per million input tokens for the cheapest, Luna, up to $5 for the priciest, Sol. Anthropic is so far holding higher prices for Opus, citing a quality edge on the hardest tasks.

For companies buying model access through APIs, these differences translate directly into bills running into thousands or millions of dollars a month at scale. More and more technical teams, including in Poland, are testing several model providers in parallel specifically so they can pick the cheapest option for a given task instead of committing to a single vendor.

xAI has not yet released detailed benchmark comparisons against Opus 4.7 or GPT-5.6 Sol, for now sticking to Musk's claims about the model's class. Independent coding benchmarks, which typically appear within days of a launch, will show whether the claimed price competitiveness comes with real code quality.

Sources: Introducing Grok 4.5 (x.ai), SpaceXAI releases Grok 4.5, which Elon describes as an Opus-class model (techcrunch.com)