I've recently cancelled Github Copilot which I found pretty good but I'm looking for an alternative. I've been looking Claude code which looks good and I'm just wondering if there are any other options to look at? I'll mainly be using it in Visual Studio Code, Xcode and Android Studio.
-
Notice regarding image hosts and attachments. See here.
Claude code or something else?
- Thread starter Cromulent
- Start date
Thank you. I guess I'll go for it.
Claude is hands down the best agentic coding tool and has been consistently for some time. Their tools generally lead the way for the competition.
Opus 4.7 and 4.8 were really step changes that made it properly good at doing all sorts of production grade code. The models were too eager to please at first and prioritised satisfying prompts over good code but they have got much better.
Fable 5 is *really* good, but frighteningly expensive. I’m making use of the limited time it’s available on Max plans to get a lot of work done. Won’t be very useful at API rates sadly, it the model must be absolutely massive.
Opus 4.7 and 4.8 were really step changes that made it properly good at doing all sorts of production grade code. The models were too eager to please at first and prioritised satisfying prompts over good code but they have got much better.
Fable 5 is *really* good, but frighteningly expensive. I’m making use of the limited time it’s available on Max plans to get a lot of work done. Won’t be very useful at API rates sadly, it the model must be absolutely massive.
I have unlimited access to Claude Code (all models bar Fable 5, for now). It's outrageous how good it is.
I've built a really complex prototype for refactoring Pivotal Cloud Foundry-dependant Spring Boot apps at large scale to Amazon ECS in about 6 hours, just using prompts. Would've taken me 4+ weeks to hand crank it out.
helps a bit 
I've built a really complex prototype for refactoring Pivotal Cloud Foundry-dependant Spring Boot apps at large scale to Amazon ECS in about 6 hours, just using prompts. Would've taken me 4+ weeks to hand crank it out.
Code:
/effort ultracode
Last edited:
Realistically, we need to get used to only AI that is useful at API rates.
There’ll always be an economic case for the fixed price subscription model with usage limits. Anthropic is expecting to turn a profit imminently so they can’t be loss-leading too badly on Claude Max.
You can run Claude code on local models. Push comes to shove you buy local hardware and run it yourself. The distilled models are getting better all the time.
Copilot's move to usage-based will signal the rest of the market. At least it'll make corporates think harder about replacing employees with AI given the substantial price difference.
Will probably get cheaper models in the future because my coding model doesn't need to know about lawn maintenance.
Associate
- Joined
- 12 May 2011
- Posts
- 490
I mean it feels natural/obvious that prices will go up, aren't most of these companies bleeding money and it's only the comical investment/hype bubble smoothing things over at the moment?
Claude is still considered the best, have been trying 'Junie' recently as it's included as part of Rider via work and it seems ok for bits and bobs, haven't drunk the kool aid of agentic coding yet so will see.
Claude is still considered the best, have been trying 'Junie' recently as it's included as part of Rider via work and it seems ok for bits and bobs, haven't drunk the kool aid of agentic coding yet so will see.
I really need to have a look at Claude code that hasn’t been setup by the company that I work for… admittedly I’ve not bothered looking at the HLD or the LLD for it.
But I just can’t get anything it gives me to work.. spent an hour with it to try and get the syntax to sed some text from the results output of another set of commands that’s been piped.. ended up within stack and getting the answer in 2 minutes.
Been speaking to a colleague at work, and he says that I’m too precise with it and I have to speak to it more generally like I would to a child.. then it will work.
I doubt if what I’m doing is that niche for Claude not to understand, but I think we (as the company) are running it on our own servers to ring fence off the data as we are highly regulated and wouldn’t want any data leaking out.
But I just can’t get anything it gives me to work.. spent an hour with it to try and get the syntax to sed some text from the results output of another set of commands that’s been piped.. ended up within stack and getting the answer in 2 minutes.
Been speaking to a colleague at work, and he says that I’m too precise with it and I have to speak to it more generally like I would to a child.. then it will work.
I doubt if what I’m doing is that niche for Claude not to understand, but I think we (as the company) are running it on our own servers to ring fence off the data as we are highly regulated and wouldn’t want any data leaking out.
The key to Claude for programming is to use Claude Code. You don't want to be looking at responses, you want it to be directly editing your code and running tests on it, etc.
You can use the chat for questions and stuff but the harness is what makes it into a really powerful tool.
I use the Claude Code CLI tool. Run it in the directory with your code it in it and set it tasks to solve where it has direct access to all your files and the build tools. It's way more effective than copy/paste. I presume you can probably also get it to work integrated in VS Code but I like to keep my tools separated. I have it set up so that I have two copies of the git project: one I get Claude to work in, the other I work on other tasks myself (bashing keys like a caveman as @mid_gen put it
) and keep an eye on what Claude is doing in the other window every now and then.Just been giving Deepseek V4 Flash a whirl, using OpenRouter and OpenCode.
Just added a new feature to my game. Followed all my existing coding standards nicely and correctly implemented the feature.
I thought it had made a mistake when I was testing it, which I asked it to investigate. It dug around the Unreal MCP for a while and discovered it was my mistake
Good first test. I think these distil models will be very much good enough for working in codebases with established guardrails like this.
Total cost $0.16.
$0.08 was the MCP work which was actually unnecessary and due to my mistake
I think I could work with this instead of Claude, if need be...and self-host is beginning to be feasible (compared to gigantic models like GLM 5.2)
Just added a new feature to my game. Followed all my existing coding standards nicely and correctly implemented the feature.
I thought it had made a mistake when I was testing it, which I asked it to investigate. It dug around the Unreal MCP for a while and discovered it was my mistake
Good first test. I think these distil models will be very much good enough for working in codebases with established guardrails like this.
Total cost $0.16.
$0.08 was the MCP work which was actually unnecessary and due to my mistake
I think I could work with this instead of Claude, if need be...and self-host is beginning to be feasible (compared to gigantic models like GLM 5.2)