Hey there,

On July 24th, Anthropic released their new Claude model, Opus 5. While this model isn’t more powerful than previous state-of-the-art models like Mythos or Fable 5, it does have significant improvements over past Opus models and is able to outperform Fable 5 in some areas. Opus 5’s main selling point is that it’s cheaper, more efficient, and Anthropic’s most trustworthy model to date. This week we’ll look at what’s changed and how this model fits into the Claude ecosystem. 

What makes Opus 5 different

Let’s start with the basics around Opus 5. This is Anthropic’s new flagship model that is meant to be used for a wide variety of tasks. This means it's the new default for Claude Max users and the most powerful model available for Claude Pro users. The new model is available for all paid plans. Opus 5 is the best choice when completing everyday tasks that involve coding and knowledge work. It is significantly better than Opus 4.8 across the board and is even able to outperform Fable 5 in some tasks. On one coding benchmark, Opus 5 more than doubles the performance of 4.8 while costing less. In one example Anthropic shared, Opus 5 on lower reasoning levels was able to match Opus 4.8 on max reasoning while using 26% fewer tokens.

What jumped out even more to me was Anthropic’s safety push. Opus 5 is the “most aligned model to date”. It directly beats Opus 4.8, Sonnet 5, and even Fable 5 by having less “misaligned” or deceptive responses. This means it's less likely to mislead you by making up information and is harder to trick into doing something it shouldn’t. Like Opus 4.8 before it, Anthropic has intentionally not trained the model in cybersecurity tasks so while it is capable of finding cybersecurity vulnerabilities, it can’t exploit them like Mythos. 

Where this fits in the bigger picture

We can’t look at Opus 5 in a vacuum. We need to look at the broader AI environment over the past few months and we can begin to see a pattern. OpenAI released their new GPT-5.6 models, Google had their new lightweight Flash models, and now we have this new Opus model that costs half as much as Fable 5. It seems like the major frontier labs have shifted their focus to creating models that give us the most capability per dollar. 

This shift makes sense when you think about it. Most users don’t need the power of a Mythos-class model, they need speed and accuracy. The high prices for both API tokens and AI subscriptions made both companies and everyday users more conscious of the price. Why pay for these more expensive models when the faster, cheaper versions are a better fit for your needs? 

The push for safer, more honest models is also clear when we look at Opus 5. Anthropic has clearly made safety a priority and I’m glad to see that they are continuing to reduce hallucinations and sycophantic behavior. They seem to be leading the frontier labs in this area and I think it sets Claude apart from other AI tools.

My Thoughts

As someone who uses Opus 4.8 on a daily basis, I’m excited it finally got an upgrade. The fact that Opus 5 performs twice as well as 4.8 in some areas means that this isn’t just a minor update but a major overhaul. I think the most important aspect of Opus 5 is that it uses fewer tokens than older versions. This means I don’t need to worry about hitting my daily and weekly quotas. One aspect of Opus 5 that I really like is the reduction in misaligned behavior. I’ve always felt that one of the main factors holding AI back is its hallucinations and deceptive behavior, so it's great to see Anthropic continue to address this issue.