Opus 4.8 Will Tank Your SEO

TL;DR: Gemini 3.5 Flash makes a strong leap to 76%, matching flagship performance from six months ago. Meanwhile, Anthropic's Claude continues a troubling downward trend. Opus 4.8 scores just 67%, reinforcing our long-standing advice: don't chase the latest release.

It’s been a minute since we updated our AI SEO Benchmark. New site, new models, new data. Let’s get into it.

Gemini 3.5 Flash: 76% (up from 69%)

First, while at Google I/O, Gemini 3.5 Flash was announced. I rushed to get access which of course it wasn’t ready yet. Let’s test Google announced Gemini 3.5 Flash at I/O, and I rushed to get access before the dust settled. The result is genuinely impressive. A 7-point jump puts a lightweight Flash model on par with where flagships sat six months ago. That’s the kind of pace that should get your attention.

I wrote more about the broader I/O announcements on Search Engine Land. Google is moving fast. Hopefully not breaking too many things in the process.

Grok: Unchanged

Exactly the same score. That’s all there is to say about that.

The Real Story – Opus 4.8

This is where it gets interesting.

Claude Opus 4.6 remains the highest-scoring model we’ve ever tested. It was praised as incredibly human and ranked as a leading model across benchmarks. I still use it every day.

Then came 4.7. Performance dropped (so much I had to crop it from the screenshot). Costs went up. Users accused Anthropic of deliberate nerfing. The company denied it, but the numbers told their own story.

Now Opus 4.8 has arrived, and the trajectory continues in the wrong direction. Opus 4.7 scored 70%. The latest 4.8 drops another 3 points to 67%.

Why? We can only speculate. IPO pressure driving cost optimization. Hyper-scaling constraints limiting their ability to serve the best models reliably to all users. Whatever the cause, the outcome is measurable and consistent: newer is not better.

Why This Matters

Our position hasn’t changed. Don’t chase the hype cycle. The latest release is not automatically the best release for your workflows. This is a lesson the industry keeps teaching, and too few teams are learning.

What worked yesterday can stop working tomorrow. That’s not a theoretical risk. It’s a practical one that shows up in your outputs, your content quality, and ultimately your pipeline.

As this technology matures and the business world builds up its immunity to release-cycle hype, the smart play is clear: move slower, test more, and rely on the people who actually operate these workflows day to day. They’ll see the changes before any press release tells you about them.