Anthropic Opus 4.7: The First Model That Disagrees With You

Añadido: 2026-05-05

[00:00:00]Anthropic have just dropped a new state-of-the-art model, Opus 4.7, and it's 50% off right now inside Augment code. Anthropic keeps pushing the capabilities of state-of-the-art coding models, and Opus 4.7 is no exception.

[00:00:13]What stands out the most is how strong it feels in long-running agent work.

[00:00:17]That matters in a lot of the workflows people care about: async task, long-running background agent work, CI/CD pipelines, and bigger jobs that run across many steps. It doesn't just work longer, it keeps the thread longer, staying with the work across many steps with less babysitting. I let it work for over 8 hours to optimize a chess engine under 1 MB to play in this online competition, and it was able to beat one of the top engines. Another thing we've noticed is that Opus 4.7 feels more opinionated. Older models could be too quick to agree. I'm sure you've seen the meme, "You're absolutely right." It keeps hyping you up, but some of the times maybe your ideas really aren't that great. Opus 4.7 actually feels like it would challenge these bad ideas. You might find it critiquing approaches and pushing back more when things don't make sense. And for real engineering work, that's a necessity. You want the model to catch a weak assumption, question a bad plan, and help steer the work in a better direction, not just one that agrees confidently and keeps going no matter what. So, between strong long-running performance and better judgment during the work, Opus 4.7 feels like a meaningful step forward. And there's no better time to try it inside Augment now that we have 50% off across all Augment products 4.7. We've just started rolling this out to all users, so you might not see it right away, but it will come within the next hour or so.

[00:01:37]Enjoy.