Claude Opus 4.8 is the strongest model available right now for large, difficult, long-running tasks, and it is the most honest and least lazy model Anthropic has released. For writing public-facing content, Opus 4.6 still does a better job. At TJ Digital, we do about 90% of our daily work in Claude across roughly 42 active campaigns, so model choice is something we test constantly.
Why Did Claude Opus 4.8 Get a Mixed Response?
The release got a mixed response, and most of that traces back to how it was built. Opus 4.8 was built on top of 4.7, and 4.7 had a mixed response of its own. A lot of people, our agency included, kept using 4.6 for most work.
The complaints about 4.7 were that it felt overly sterile, it took everything literally instead of applying common sense judgment, and it used a lot more tokens to get to an answer.
Opus 4.8 is picking up many of those same criticisms. Anthropic only published the benchmarks where the model performs well, which is normal, but it does perform well on those benchmarks.
The headline strengths are real. It is currently the most honest and least lazy model, and Anthropic specifically designed it for long-running tasks. So far that is exactly where it shines.
What Is Claude Opus 4.8 Best For?
If you have a very large or difficult project that you want a model to grind through, and you do not mind spending a lot of tokens, Opus 4.8 might be the best option available today. Think long code migrations, deep analysis, and multi-step work where getting it wrong is expensive.
You also get full control over how hard it works through an effort setting. The available levels are:
| Effort Setting | When to Use It |
| Low / Medium | Not recommended for serious work. Results are unreliable below high. |
| High | The reliable baseline for real tasks. |
| Extra (XHigh) | Tough or high-stakes projects worth the extra tokens. |
| Max | The heaviest jobs where correctness matters most. |
| Ultra Code | A setting inside Claude Code for massive codebase work. |
In my early testing, anything below high produces unreliable results. At that point you might as well use 4.6. So if you are reaching for 4.8, keep the effort on high or above, or you are not getting the value out of it.
Should You Use Claude Opus 4.8 for Content Writing?
No. If you are writing any kind of public-facing content, I would stick with 4.6 for now. The same sterile, overly literal quality that holds back 4.7 and 4.8 shows up most in writing, where warmth and judgment matter most. That call lines up with how we pick the right model for each job in our guide to AI for work.
There is real concern in the community that Anthropic will retire Opus 4.6. It was already removed from the desktop app. Hopefully Anthropic saw the response and decided to keep it around for a while longer.
Will Claude Opus 4.8 Stay the Smartest Model?
Probably not for long. In the same announcement where Opus 4.8 launched, Anthropic said a new class of model was coming in the following weeks, and we know that to be the Mythos class. Anthropic’s own Opus 4.8 announcement points to stronger safeguards being needed for models at that capability level.
We can expect Mythos to be very good at coding and very expensive. I broke down what is coming next in my piece on the new ChatGPT and Claude models.
It seems like Anthropic is currently all in on building the most powerful models for software engineering. That makes sense, because it is where their money is coming from, and it lets them use their own models internally to develop their products.
The risk is that it leaves the door open for OpenAI to make Codex the go-to product for knowledge work.
How Should Businesses Handle All These Model Releases?
The practical answer for now is to stay agile. Models are changing fast, and the smartest one this month may not be the one you reach for next month. The businesses that adapt fastest are the ones already running about four times the work at the same cost by building their operations around these tools.
The most important thing you can do is structure your data and your workflows so you can swap out the model whenever a better one comes along. If your systems are tied to one specific model, every release becomes a scramble. If they are not, every release is just an upgrade.We build AI-powered marketing systems designed to keep pace with releases like this. Reach out to TJ Digital and we will take a look at where you stand.