Elon Musk's xAI shipped a new flagship model on Wednesday, and the headline isn't the benchmarks. It's the price.
What xAI launched
Grok 4.5 is xAI's newest model, and the company says it's the first one it built specifically for coding and agent work. It was trained alongside Cursor, the AI coding tool SpaceX recently acquired, and it's live today in Grok Build, in Cursor on every plan, and through xAI's API. One catch: it isn't available in the EU until mid-July.
Musk framed it plainly: "It is an Opus-class model, but faster, more token-efficient and lower cost." In a follow-up he called it "roughly comparable to Opus 4.7, but much faster."
The pitch is the price
This is where Grok 4.5 tries to stand out. It's priced at $2 per million input tokens and $6 per million output tokens. For comparison, Claude Opus 4.8 runs $5 and $25, and OpenAI's GPT-5.6 runs $5 and $30. xAI also says the model solves tasks using far fewer steps, which lowers the real cost per job even further. And for a limited window, it's free to use inside Grok Build and Cursor.
Speed is the other half of the sell. xAI serves Grok 4.5 at around 80 tokens per second, which it describes as faster than typical "flash" speed models, so answers come back quickly.
It's aimed at your desk work too
Grok 4.5 is now the default model in Grok Build, and xAI leaned hard on office tasks in the launch. It can build multi-sheet Excel models that pull in research from the web, assemble PowerPoint decks using native shapes and diagrams, and write clean Word documents, all from a prompt. There are plugins for Word, PowerPoint, and Excel to wire it into the apps directly.
How good is it, really?
The independent picture is more mixed than the marketing. On xAI's own benchmark charts, Grok 4.5 often lands just behind Anthropic's Fable 5 and OpenAI's GPT-5.5 on coding tasks, roughly in Opus territory rather than clearly ahead of it. One design-focused tester, BridgeMind, said Grok 4.5's UI work still trails Claude Opus. Others were more enthusiastic: developer Theo called it "pretty damn good and REALLY well priced," and a Website Arena leaderboard placed it fifth, a 25-rank jump over xAI's previous model. The honest read: it's a real step up for xAI and a genuine bargain, not a new king.
I haven't tested this myself yet. But if it really lands just behind Fable 5, even if that's only xAI's own benchmark, it's probably worth a test. I'll report back.
And at that kind of cost, it might be worth just using day to day. It's also a good one to route to in a bigger setup: let Fable 5 orchestrate and fan work out to Grok Build and GPT-5.6.
If you're already in Cursor, it's nice that it's built right in too.