Anthropic Launches Claude Opus 4.5 With Improved Coding and Agent Capabilities

I don't use AI and you shouldn't either.

Opus 4.5 is awesome and the usage limits are now finally generous. There is some great detail about what they've done and the videos for this release are really worth watching, especially the one comparing Sonnet 4.5 vs. Opus 4.5 on context usage and problem solving / opening 'locks'.

If you used Claude Code, compacting now works in chats automatically.

I'm not sure if this is good or not for very long context, I tend to prefer summarizing at the end and carrying that forward manually, I imagine the context window is sliding once the compaction happens so you probably won't be able to get a real summary like you would before, but for people who change topics or keep a long thread open it will probably be a welcome change. I'd like this toggle-able per chat but I doubt that will happen.

Max users get signiifcantly more Opus usage than before, as much as Sonnet previously. Opus is the default model for me for the first time ever in the UI which is great. It even defaulted to extended thinking (!).

I'm in the middle of a 3 hour session now and have used les than 1% of my week; easily 1/3+ less than usual. VERY happy with this from Anthropic and can't wait to use Opus 4.5 with WebStorm in a few weeks to refactor some typescript.

What a contrast (at least for me). Even just from the articles here on MR, with seemingly every update, OpenAI is so focused on making its models blow more smoke up your ass and getting involved in commerce, while Anthropic is over here, by all accounts, making its models better at producing code (one of the only things I use these tools for) and following prompts while raising usage limits. Really looking forward to trying this out over the next week.

Now, one issue was that I hit the usage limit on the free tier relatively fast, anyone know if the Pro plan is (considerably) more generous, the pricing page just says: More usage*

If you want to get things done with ANY AI tool, don't use the free tier. You have to pay to play right now. If you're doing even medium scale development and using them regularly you have to pay for Max, not Plus. Might not be worth it to you; I've saved easily $40k in developer costs in the last week alone and it cost me about $1,000 in usage.

For trying it out a bit more, I'd recommend you give Plus a shot and see if it meets your needs. They do pro-rate you if you upgrade.

I agree with you that AI is here to stay and will replace jobs. But you overestimate the benefits of being early. AI tools are still in their infancy but they improve incredibly fast. It is not necessary to become "experienced", as this won't be required at all in just a few iterations. It is as if you were the first to adopt driving a 19th century car with profound mechanic skills, abilities to frequently repair the engine, drive, park etc. And two years later the whole technology is already replaced by self-driving cars that do everything voice-controlled and without getting your hands dirty.

Self driving cars are actually a great analogy, because they work under similar technological hopes and dreams yet you still need to know how to drive a car unless you're living in a specific city where everything is mapped. So I think your analogy is somewhat apt but your conclusion is a little skewed.

The person who will keep their job is the one with the best business domain knowledge, analyst and communication skills, who can therefore achieve he most with whatever limitations AI will still have. Not the one guy who tinkered with every model since GPT-2. Everyone will just use similar tools in the same ways.

I don't think this is true, it won't be dumbed down enough – for general use, yes, like office tasks etc. For programming and architecting you will need to understand the domain and the technology. Using the models now can help you get up to speed for the future that's coming, I absolutely do not think it's a waste of time and the main caution I'd have is relying too much on the output and not understanding what is actually occurring behind the scenes or with the technology being implemented. e.g. if an AI writes some module, it behooves you to really dive in and learn what that module does and why it's structured that way – and press on it (with or without AI's help) to understand it and also how to improve it yourself.

As I said in another thread, "coders" are not "Computer Scientists" and that difference is going to become enormously stark very quickly. Anyone in this field should be sprinting to catch up unless they are close to retirement because the luxury salaries and jobs are going away for the regular coders.

Specifically on your point about "analyst and communication" you are 100% correct. This is critical and it's something programmers suck at. Similarly, most business people also suck at programming. The rare person who bridges those gaps and has a good high level knowledge but can understand when something doesn't make sense are going to 10x their output and improve their product quality, but it won't be the norm.


A curious person who bridges the gap in a very similar way to "the intersection of liberal arts and technology" that Steve Jobs talked about... we're exactly at the bicycle for the mind part of all of this and it is thrilling. The number of people dumping on AI who do not understand it just further proves that willful ignorance is a major problem, despite there being objective issues with the technology from ethics to sociological impacts to societal surveillance and control. The thing is, that stuff is happening whether we want it to or not.

If this is someone's career (not you specifically) and they're early or even later mid-stage and ignoring AI and they don't work on a very narrow domain that is extremely uncommon (RTOS, HPC, Compilers, PhD frontier research) their trajectory is going to get absolutely obliterated if they don't learn and learn fast.


edit: the cowards who just post a laugh emoji and don't engage in discussion are hilarious :)

Hmm... did uou know that everything you are using - your phone, computer, home tech, etc. - use some form of AI? Perhaps you shouldn't even be talking on this board either.

People are afraid of AI taking their jobs so they bury their heads in the sand hoping that this is just a bubble that will collapse soon and they can go back to pre-2022.

The only people who won’t be employed and won’t be able to compete are the ones who don’t use AI to assist them.

OpenAI has close to 1 billion users, I’m sure most of them are not interesting in coding.

I don't doubt that in the slightest. Doesn't mean I like what the company is doing.