SpaceXAI's new Grok 4.5 model costs $6 per million output tokens, compared to $30 for OpenAI's GPT-5.6 and $25 for Anthropic's Claude Opus 4.8, making it one of the cheapest frontier-class AI models available. But the aggressive pricing comes with a trade-off: independent testing shows Grok 4.5 ranks fourth on capability benchmarks, not first, and carries a 54% hallucination rate on factual questions.

The launch on July 9, 2026, marks the first major release under SpaceXAI's new corporate identity, following the company's rebrand from xAI earlier in 2026. Elon Musk framed Grok 4.5 as "Opus-class, much faster," comparing it to Anthropic's top model. However, that vendor claim does not align with independent measurements from Artificial Analysis, a neutral benchmarking service.

How Are AI Companies Competing on Cost Right Now?

The AI industry is undergoing a dramatic shift in pricing strategy. For the past year, companies encouraged businesses to use AI as much as possible in a practice called "tokenmaxxing." But as enterprises received sticker shock from usage-based billing, developers began racing to offer more efficient models at lower prices.

• OpenAI's Response: GPT-5.6 is designed to complete more work while using significantly fewer tokens, a unit of data processed by AI models. The company also introduced credit usage analytics and spending controls in June to help customers manage costs.
• Meta's Strategy: CEO Mark Zuckerberg said Meta is prepared to be "aggressive" on pricing, noting that "the pricing from some of the other labs is very extreme and has very high margins." Meta's Muse Spark 1.1 is positioned as a high-level model at much more affordable cost.
• SpaceXAI's Bet: Grok 4.5 lists at $2 per million input tokens and $6 per million output tokens, roughly half what Anthropic charges for Claude Opus 4.8 and a quarter of GPT-5.6's output token price.

"Every enterprise now is thinking about spend and the value they're getting in exchange for AI, and this is what we really want to do," said Sam Altman, CEO of OpenAI.

Sam Altman, CEO at OpenAI

One Paris-based AI startup CEO told researchers that he had spoken with executives whose monthly AI bills reached millions of dollars after using models from OpenAI and Anthropic. Gil Luria, head of technology research at DA Davidson & Co., noted that as costs spiral, companies are "starting to ask questions about efficiency".

Where Does Grok 4.5 Actually Rank on Capability Tests?

Grok 4.5 scores 54 on Artificial Analysis's Intelligence Index, placing it fourth behind Claude Fable 5, GPT-5.5, and Claude Opus 4.8. The model's strongest performance comes in coding tasks, where it scores 76 on the Coding Agent Index, roughly matching GPT-5.5 while costing far less to run.

For agentic coding work, where a model processes large numbers of tokens across many steps, the combination of GPT-5.5-level coding output with Grok 4.5-level pricing creates a compelling case for engineering teams to test the model. On a per-task basis, Artificial Analysis estimates Grok 4.5 costs $2.49 to complete its standard evaluation, compared to $11.80 for Claude Fable 5, the current index leader.

However, accuracy on factual questions is a significant weakness. Artificial Analysis recorded a 52% accuracy rate and a 54% hallucination rate on its AA-Omniscience test, meaning the model confidently invents answers more than half the time when it gets a question wrong. This trade-off is typical for fast, cheap models and suggests Grok 4.5 works best in workflows that verify their own outputs, such as code that runs or retrieval systems that cite sources.

The model ships with a 500,000-token context window, allowing it to process roughly 400,000 words at once. It supports three reasoning levels (low, medium, and high) and accepts both text and image inputs.

Why Is Grok 4.5 Blocked in Europe?

At launch, Grok 4.5 is unavailable in the European Union under the EU AI Act's systemic-risk rules, which restrict the deployment of the most advanced AI models without additional safeguards. SpaceXAI has framed this as a staggered rollout rather than a permanent ban, signaling EU availability around mid-July 2026.

This regulatory hurdle means an entire regulated market cannot access SpaceXAI's flagship model at launch, a constraint that does not affect US-based competitors in the same way. The timing suggests SpaceXAI is working to meet EU compliance requirements before full rollout.

What Does This Mean for the Broader AI Market?

Grok 4.5's aggressive pricing puts pressure on Anthropic, which many viewed as the market leader. Anthropic's Opus and Fable models rank among the most expensive on a cost-per-task basis, according to Artificial Analysis data. Musk specifically targeted Anthropic in a post touting Grok 4.5, describing it as "an Opus-class model, but faster, more token-efficient and lower cost".

The broader shift toward cost efficiency reflects real business pressure. Chinese tech companies like DeepSeek have flooded the market with affordable open-source AI models. While these services still lag behind cutting-edge US options, they are good enough for many routine tasks. Some users are turning to model routing services, which allow them to select from hundreds of AI models for different tasks to ensure better prices. OpenRouter, one such service, raised more than $100 million in funding in May to meet demand.

The pricing war also signals a departure from earlier rhetoric. About a year ago, OpenAI executives publicly mused about charging thousands of dollars per month for top-tier AI models to reflect their growing value. Now, the conversation has shifted entirely to cost management and efficiency.