Grok 4.5 undercuts Claude Opus on price but trails it at the top

SpaceXAI recently released a new version of its AI chatbot called Grok 4.5. This is a coding-focused model that Elon Musk pitched as an “Opus-class” rival to Anthropic’s Claude at a fraction of the cost. 

Considering the rising cost of AI usage, engineering teams are now prioritizing price per completed task over peak intelligence. 

Does Grok’s new model outperform Claude Opus 4.8?

Elon Musk, in a post on X, described the newly launched Grok 4.5 as “an Opus-class model” that is faster and more cost-efficient. The model is designed for coding and other complex tasks. 

Musk followed up after the launch with a second post, writing, “Looks like Grok 4.5 is #1 on at least a few benchmarks. Better than expected.”

SpaceXAI (NASDAQ: SPCX) lists Grok 4.5 at $2 per million input tokens and $6 per million output tokens, which is significantly cheaper than using Anthropic’s models, which are priced at $5 and $25 for Claude Opus 4.8, or OpenAI’s GPT-5.6 Luna, which sits at $1 and $6. 

The new model is the first to be developed jointly with Cursor, the AI coding startup that SpaceXAI recently agreed to buy for $60 billion in an all-stock deal. The deal is expected to close in the third quarter of 2026. 

Cursor said it partnered with SpaceXAI on training, feeding real signals on how engineers write, review, and debug. The training ran across tens of thousands of Nvidia GB300 GPUs, and SpaceXAI says the model runs at about 80 tokens per second with roughly twice the token efficiency of comparable rivals.

Where does Grok actually rank among its rivals? 

The independent benchmarking site, Artificial Analysis, scored Grok 4.5 at 54 on its Intelligence Index, placing it fourth behind Anthropic’s Fable 5, OpenAI’s GPT-5.5, and Claude Opus 4.8. 

In coding tests, Artificial Analysis gave Grok 4.5 in the Grok Build a score of 76 on its Coding Agent Index. This is at the same level as GPT-5.5 in Codex and just under Fable 5 in Claude Code. 

The site measured Grok 4.5 at $2.49 per coding task compared to $11.80 for Fable 5 in Claude Code. Grok 4.5 burned through roughly 1.9 million tokens per task, whereas Fable 5 used 7.2 million.

On one specific benchmark, Grok 4.5 has a clean win over Claude. Snorkel AI ran the model against GPT 5.5 and Opus 4.8 on a sample of about 2,000 professional-work tasks from its GDPval+ dataset. 

Grok 4.5 passed 29% of them, ahead of GPT 5.5 at 22% and Opus 4.8 at 21%. The widest margins were in legal, education, and healthcare tasks. Grok 4.5 was paired with SpaceXAI’s own Grok Build agent for that test at the company’s request.

Cryptopolitan reported earlier in the week that Elon Musk’s statement of the model being “close to, perhaps exceeding Opus” originated as a result of the impression from company evaluators rather than a published benchmark. The specific Opus version Grok was measured against was Claude Opus 4.8. 

Musk himself later conceded that the model competes with a prior Claude generation rather than the current one.

Grok 4.5 is available now through the SpaceXAI console, the Grok Build agent, and inside Cursor on all plans, with EU access expected in mid-July. 

If you're reading this, you’re already ahead. Stay there with our newsletter.