Hello folks! There’s a new AI model out today, and it’s shaking up the leaderboard. xAI (Elon Musk’s AI company, now folded into SpaceXAI) just released Grok 4.5, and the numbers are getting attention for a reason nobody expected: it’s not the smartest model out there, but it might be the smartest one for the price.
What actually happened
xAI shipped Grok 4.5 on July 8, and independent benchmarking firm Artificial Analysis ran it through its standard tests within a day. The result: a score of 54 on the Intelligence Index, good for fourth place overall. That puts it behind Claude Fable 5 at number one, GPT-5.5 at number two, and Claude Opus 4.8 at number three — but ahead of every open-weight model and notably ahead of Google’s Gemini line. You can see the full breakdown on Artificial Analysis’s benchmark writeup.
That 54 is a real jump. Grok 4.5 gained 16 points over its predecessor, Grok 4.3, which is one of the bigger single-version leaps we’ve seen this year. Under the hood, the model reportedly runs on around 1.5 trillion parameters, about three times the size of the previous version, with a 500,000-token context window (that’s roughly the length of a long novel it can “read” at once).
The part everyone’s actually talking about: the price
Here’s the detail that turned this into a bigger story than a normal model launch. Grok 4.5 costs about $0.31 to complete a typical Intelligence Index task. For a coding task specifically, Artificial Analysis clocked it at roughly $2.49 — compared to $11.80 for Claude Fable 5 doing the same job. Token pricing sits at $2 per million input tokens and $6 per million output tokens, with a 75% discount on cached tokens.
Part of the reason it’s so cheap to run is efficiency, not just a lower price tag. On coding benchmarks, Grok 4.5 reportedly used about 1.9 million tokens to finish a task that took competing models 6.2 to 7.2 million tokens. Fewer tokens burned means a smaller bill, even before you factor in the per-token price.
On coding specifically, Grok 4.5 landed third on the Artificial Analysis Coding Agent Index, roughly on par with GPT-5.5 running in Codex, though still behind Fable 5. It also posted the top score on a banking-agent benchmark called τ³-Banking, beating GPT-5.5 there.
Why this matters if you’re not a developer
You might not touch Grok directly, but this kind of price war affects you indirectly. Companies building AI features into apps you already use — customer service bots, coding assistants, research tools — pick models based on both quality and cost. When a model this capable shows up this cheap, it pushes the whole market toward lower prices and faster, more affordable AI features for everyone downstream.
If you do use AI chatbots for work — and if you’ve tried OpenAI’s new ChatGPT Work agent, you already know how fast this space moves — Grok 4.5 is available now through xAI’s apps and API for anyone who wants to try it directly. I ran a few coding prompts through it this morning and it felt noticeably snappier than earlier Grok versions, though I’d still reach for Fable 5 or GPT-5.5 first for anything I need to get exactly right on the first try.
What’s next
Google’s Gemini 3.5 Pro is still MIA in this race. It’s now six weeks past its original June 30 target and hasn’t shown up outside a limited Vertex AI enterprise preview, so it’s conspicuously absent from these rankings. Expect more head-to-head comparisons as Gemini eventually ships, and don’t be surprised if OpenAI or Anthropic respond to Grok’s price pressure with their own cheaper tiers before the year is out.
Frequently asked questions
Is Grok 4.5 free to use?
No. It’s available through xAI’s paid apps and API, with usage-based pricing per token. Some access may come bundled with X (formerly Twitter) Premium tiers, but check xAI’s site for current plans since pricing details change often.
Is Grok 4.5 better than ChatGPT or Claude?
Not on raw intelligence score — it ranks fourth, behind Claude Fable 5, GPT-5.5, and Claude Opus 4.8. Where it wins is cost efficiency: similar coding results for a fraction of the price.
What does “Intelligence Index” mean?
It’s a composite benchmark run by Artificial Analysis, an independent AI testing firm, that combines several reasoning, coding, and knowledge tests into one score so models can be compared side by side.
Who owns Grok now?
xAI, Elon Musk’s AI company, which has been increasingly referred to as SpaceXAI following closer integration between the two companies.
Would you trust a cheaper AI model for daily work, or do you stick with the priciest option because it’s “safer”? Let me know in the comments.