On Friday I mentioned I was going to write a review of Opus 5.5. This is it. I'll be posting a more technical version on X for engineers. If you want that one, go follow me there.
The short version
Every model I've ever used has come with a compromise.
GPT-6 Astra was incredible at driving a computer, but it tended to fall apart on random tasks.
Fable 5.1 was amazing across the board, but incredibly expensive. I bought six Anthropic plans just to keep using it, and I still kept running out of tokens. It was worth it. It was running a lot of my business. I tried moving that work to Astra. It was close, but it just didn't work quite as well.
GPT-5.6 was amazing because it wouldn't stop until the task was done. But sometimes it got so focused on the task that it destroyed things along the way.
Opus 5.5 might be the first model I've ever used with zero compromises. It's genuinely the best model I've ever used. It feels like Anthropic messed up and gave me five times more for my money. This model can do so much and barely make a dent in my subscription plan. It's also faster, way faster than Fable. In my experience, there is zero downside to switching from Fable to it.
What it's like to use
It's stable. The current OpenAI models are brilliant on some things and kind of fall apart on others. Fable 5.1, on the other hand, was just stable, and Opus 5.5 has the same feeling. It just works.
It's also easier to talk to. Previous Anthropic models slip into "Claude-speak", where the answer reads like gibberish because the model is so far ahead of you that it stops explaining itself. This one doesn't do that at all. It's a great communicator. And it's responsive, reliable and proactive.
It has flaws, and they're small. Sometimes it overestimates how long a job will take and says, "I'll have this to you by tomorrow morning." I say, "Just do it now," and it says, "Okay, 5 minutes." It's great at driving your computer, though OpenAI's Codex app is still a little smoother at it than the Claude apps, and part of that gap is Claude's app, not the model. Next to the catches above, these are nits.
What it unlocks
Because it's faster and cheaper, you can give it things you couldn't before. I'm able to put way more AIs, and way more work, on the problems I'm facing.
I've been using it for normal work and it's absolutely amazing, but I've been pushing it on even more interesting things just to see what it can do. For example: a rebrand for Something Big, and a New York City open-world game. I even bought a 3D printer and I'm having Opus learn to use it. Maybe I'll make a video about this pretty soon. These are things that previous models couldn't touch. Obviously this isn't useful in daily work, but it gives you a sense of just how capable it is.
I'm running so many of these at once that I needed a new computer to keep up. The agents picked out the one they needed and ordered it for me.
They're also getting insanely good at communication. I'm now in group chats with team members, partners and agencies, and all of us have Opus 5.5 agents in there, talking to each other to get work done all day. More and more of it happens without us. The work is good, too. Really good.
I haven't touched my computer since this model came out. I text my agent what I want done, and it does it. That's a custom setup, so don't expect it out of the box, but it should give you a sense of where things are going. If you're interested in reading about how I set this up, please reply and let me know!
It won't be the best for long
I think this model will be outdated pretty soon. From what I'm seeing and hearing, the next set of models will make this one look like a toy. But really, if you liked Fable, you'll love this.
What to do
- If you're on a paid Claude plan, open Claude and make sure Opus 5.5 is the model selected.
- If you moved work off Fable because of cost, to Astra or anywhere else, move it to Opus 5.5.
- Pick one job you'd written off as too expensive for AI, and try it this week.
If you try it, hit reply and tell me what you gave it. I can't reply to everyone, but I read every single one.
— Matt