xAI has a bold pitch for Grok 4.5: it codes about as well as Claude Opus 4.8, but does it with roughly a quarter of the tokens. That’s not a “we’re smarter” claim. It’s a “we’re just as good for far less money” claim, which in 2026 might matter more. Someone actually put it to the test, so let me walk through what the numbers say and why you should care.
The claim and the pricing
Grok 4.5 landed on July 8, 2026. xAI says it matches Opus 4.8 on coding while using about 4.2 times fewer output tokens to get there.
The sticker price already favors Grok. It runs $2 per million input tokens and $6 per million output. Opus sits at $5 and $25. That’s less than half the price on both sides before you even factor in the token efficiency. Stack the two together and the cost gap gets dramatic.
What the benchmarks say
The benchmarks mostly support the marketing, with a catch. On Terminal-Bench 2.1, which measures real command-line work, Grok 4.5 scored 83.3 percent to Opus 4.8’s 78.9. But on SWE-Bench Pro, the harder test of fixing real open-source bugs, Opus still comes out ahead.
So the honest read is not “Grok is better.” It’s “Grok is about as good, for a lot less.” Different claim, and a more interesting one.
The hands-on test
Benchmarks are one thing, real work is another. The New Stack ran a head-to-head, giving both models the same three jobs in one real Rust project (the fd file-finder) inside Cursor, and tracked every token. I’m summarizing their results here, credit to them for actually measuring it.
The three tasks were a bug fix, a multi-file refactor, and a feature build. The code both models produced was nearly interchangeable, so the story came down to tokens, time, and cost.
On the small bug fix, Opus actually won. Both wrote an identical fix with all tests passing, but Opus did it faster and on fewer tokens. Grok’s efficiency edge showed up on the bigger jobs. On the refactor, Grok used about 197K tokens versus Opus’s 954K for the same result, roughly a fifth. On the feature build, Grok spent about 603K tokens to Opus’s 3.2 million, with Opus only slightly more thorough (it wrote the man page entry too).
Add it up and the totals are striking. Across all three jobs, the reported numbers were about 1.01 million tokens and $1.00 for Grok, against 4.33 million tokens and $5.14 for Opus. That’s Opus using 4.3 times the tokens, almost exactly the gap xAI advertised.
The caveats worth knowing
Before you cancel anything, the test author was upfront about the fine print, and it matters.
Those were Cursor’s blended token counts, not raw API numbers, so they include context each agent re-sends every turn. Grok ran on Cursor’s fast tier while Opus ran on its thinking tier, so some of Opus’s token pile is extended reasoning, not pure waste. And Grok’s price had a 50 percent promo, though even at full rate its three jobs came to about $2 versus Opus’s $5.14. It’s also three tasks in one repo, not a giant benchmark. Small sample, real signal.
What it actually means for developers
Here’s the part I care about, because it’s a cost story dressed as a model story.
For a lot of everyday coding, these two models are interchangeable on output. If the code comes out the same, the tiebreaker becomes tokens, time, and dollars, and on the bigger jobs Grok won those clearly. When you’re doing something at scale, hundreds of tasks instead of three, a 4x token gap turns into a serious line item.
That said, Opus wasn’t beaten on quality. It was a shade more thorough on documentation and it won the small, fast task. If your work leans toward hard bug-fixing or you value that extra polish, the premium may still be worth it.
The bigger shift is that “just use the most powerful model for everything” is getting expensive to justify. The smart move now is to match the model to the task and watch the cost per job, not the brand on the box.
The bottom line
Grok 4.5 didn’t prove it’s better than Claude Opus 4.8. It proved it can do a lot of the same coding work for roughly a quarter of the tokens and a fraction of the price. For teams past the stage of not caring what AI costs, that’s the whole ballgame. Test both on your own workload, track the tokens, and let the bill help pick the model. As always, benchmarks start the conversation, your repo finishes it.
답글 남기기